A Comprehensive Survey on Federated Learning Applications in Computational Mental Healthcare

Abbreviation

1  Introduction

Mental health is vital for good wellness, even though diagnosis and treatment of mental illnesses are difficult due to their varied complexity. In particular, the application of technology in mental health service provision has undergone considerable improvement in recent times [1,2]. Collaborative machine learning, known as federated learning, can be promising for complex mental health applications [3]. This research will measure and summarize the current landscape of mental health applications in FL.

Typically, in traditional models, mental health centers train models apart within the settings, as illustrated in Fig. 1. Instead, federated mental health learning is a revolutionary strategy [4]. Fig. 2 entails training individual models at different centers and updating the single centralized model with these models. Importantly, this process occurs without sharing raw data, marking a significant shift in privacy preservation and collaborative learning strategies [5].

Figure 1: A traditional learning for mental health application

Figure 2: The federated learning of mental health application

This research focuses on federated learning approaches ML and DL in the mental health field to enhance knowledge, diagnosis, and cure of mental health diseases.

This study compares cutting-edge federated learning apps and traditional mental health apps. It aims to conduct a brief analysis focusing on the various facets that define this dynamic industry. By extensively investigating the datasets used, the data sources supporting these apps, and the specific mental health symptoms addressed, this study aims to assess the healthcare industries.

The sources and data types are discussed at the outset of this article’s structure. This research investigates federated mental health applications that are categorized based on symptoms. Next, it details how it can theoretically incorporate Federated learning ideas into the proposed Federated mental health system. The processes involved in federated deep learning and machine learning are explored. The following sections describe the advantages of mental health apps. The study concludes by outlining possible directions for the future and providing a holistic view of development for federated learning in mental health applications. This research aims to advance the ongoing discussion over technology compatibility and mental health.

The study’s contributions are presented as follows:

•   Introduction of Federated Learning in Mental Health: Federated learning embracing complex mental health issues.

•   Revolutionary Collaborative Learning Strategy: Supporting a move from the conventional individual sample training model to group cross-center training, maintaining privacy and not sharing raw data.

•   Comparative analysis of federated learning applications and traditional mental health apps, datasets, data sources, and targeted mental health symptoms.

•   In-depth Exploration of Processes and Advantages: Exploring the mechanisms behind federated deep learning and machine learning that provide an understanding of how they improve the knowledge, diagnosis, and treatment of mental health diseases.

•   The Comparative Study of Federated Learning Surveys in Healthcare

Table 1 lists several survey papers that explored federated learning in healthcare, and they are viewed from numerous perspectives, such as general healthcare, IoT, EHR, and blockchain, but no work has integrated mental health aspects, which are essential these days. Therefore, this study aims to fill the gap in reviewing the research on federated learning in mental health applications.

2  Dataset Source for Mental Health Apps in Federated Learning

With a focus on federated learning for mental health applications, diverse data sources are the foundation for understanding, analyzing, and improving mental well-being, as shown in Fig. 3. These sources include:

Figure 3: Dataset source information

2.1 Mobile Health (mHealth)

Federated learning for mental health applications depends on mobile health (mHealth) as a data source. People freely provide real-time mental health data through mHealth apps on their smartphones and tablets, providing insightful information while maintaining the privacy and confidentiality of their sensitive data [14,15]. Federated learning integrates data from several mHealth sources to enable cooperative model training for enhanced mental health therapies [16–19].

2.2 Survey Data

Participant surveys are essential to gathering structured data for federated learning for mental health. Machine learning techniques from federated learning can be employed to assess mental health behavior, attitude, and symptom survey data. This ensures that the opinions of various respondents are combined to develop a detailed and protected mental health model [20–23].

2.3 Wearable Devices

Devices like smartwatches and fitness trackers can be dynamic physiological data sources for federated mental health learning [24–27]. They protect individual user privacy, returning vs. open-ended monitoring of mental-well-being-linked data. Federated learning utilizes data from multiple wearables to support the training of models in a cooperative manner, delivering personalized insights without centralizing sensitive data [28–30].

2.4 Electronic Health Records (EHR)

The wide-ranging data sources of electronic health records significantly support federated learning-based mental health applications. Federated learning helps to integrate knowledge from different EHR data sources for a more inclusive understanding of a person’s treatment plans and mental health history. Therefore, this strategy promotes precision in mental health interventions [31–34].

2.5 Social Media Contributions

Social media sites as sources of mental health data in federated learning settings. Trends and attitudes in mental health via social media conversations, knowledge exchange, and community involvement. For example, federated learning enhances the knowledge of mental health in diverse populations by incorporating their contributions collectively and guaranteeing user’s privacy [35–38].

2.6 Experimental Data

Experimental data refers to the structured and controlled data generated from clinical trials, laboratory studies, or simulated environments specifically designed to study mental health interventions. Regarding federated learning, experimental data can be used for model training on complex scenarios to capture through mHealth apps, surveys, wearables, or EHRs. This data can be invaluable for testing hypotheses, validating results from other sources, and ensuring that federated learning models can generalize across different conditions and settings. By integrating experimental data, federated learning can contribute to a more robust and comprehensive understanding of mental health, ensuring that models reflect real-world data and are grounded in scientific research. This data is essential for fine-tuning models and validating outcomes from various federated learning processes [39–42].

3  Dataset Type for Mental Health Apps in Federated Learning

Applications related to Federated Learning for Mental Health use many datasets, each offering an individual insight into the complexities of mental health. These datasets are divided into several indications of different methods for understanding and diagnosing mental health, as depicted in Fig. 4.

Figure 4: Dataset types information

3.1 Text-Based Datasets

Mental health applications should involve text data such as user-generated content, clinical notes, and textual indications of emotional states. These include text data for sentiment analysis, pattern detection, and relevancy information extraction using natural language processing techniques [43–46]. Fig. 5 illustrates the process of utilizing textual data in a sentiment analysis task.

Figure 5: Text data processing using machine learning models for predicting sentiment analysis

3.2 Audio and Speech Datasets

Mental health applications gain auditory dimensions from speech and audio datasets. Tone detection, voice analysis, and speech patterns are helpful ways to determine someone’s emotional state. These datasets enable the creation of algorithms that identify and respond to emotional cues in spoken communication [47–50]. Fig. 6 depicts the process of speech signal processing for depression identification.

Figure 6: Speech signal processing using deep learning models for detecting depression state

3.3 Signal Datasets

For example, EEG provides neurological signals when studying brain correlates of mental health. Federated learning and EEG datasets can generate applications that read thoughts to decipher emotions, stress, and cognitive status [51–54]. Fig. 7 shows the emotion recognition processing using EEG signals.

Figure 7: EEG processing based on deep learning models for emotion recognition

3.4 Image Datasets

Mood-related imagery and facial cues in the visual representation provide consequential perceptual context for interpreting mental health visually. Image derivate face signals and other visual cues are utilized in the normalizing models that analyze emotional well-being under federated learning applications [55–57]. Fig. 8 presents the image processing for predicting facial depression.

Figure 8: Image-based using machine learning models for predicting facial expression

3.5 Multimodal Datasets

Multimodal datasets include text, voice, signals, and performances, but a few result in rich mental health information. Federated learning applied to multimodal datasets incorporates a holistic approach that improves the accuracy and inclusiveness of mental health assessments utilizing data from multiple sources [40,58–60].

Applying multiple dataset types in Federated Learning-Based Mental Health Apps guarantees a multi-dimensional perception of people’s mental health, which assists in creating efficient and personalized treatments. Fig. 9 shows the multimodal data processing.

Figure 9: Multimodal data processing

3.6 The Frequently Used Datasets

Table 2 lists seven frequently used datasets by dataset type and task.

3.6.1 DAIC-WOZ Dataset

The DAIC-WOZ (Distress Analysis Interview Corpus) is a clinical interview database designed to facilitate the exploration of mental health issues, such as anxiety, depression, and PTSD.

3.6.2 WESAD Dataset

The WESAD (Wearable Stress and Affect Detection) dataset is a valuable resource for affect recognition, specifically focusing on wearable stress monitoring systems to enhance human-computer interaction. The dataset addresses the gap in the affective computing community by providing a publicly accessible, multimodal dataset of reliable data for affect detection and stress analysis.

WESAD includes motion and physiological data collected from both chest-worn and wrist-worn devices. It comprises body temperature measurements, electrodermal activity, electrocardiogram, electromyogram, respiration, and blood volume pulse.

This dataset goes beyond previous studies by combining three affective states, stress, neutral, and amusement, providing a better understanding of emotional states.

The dataset also incorporates self-reports from 15 subjects gathered through established questionnaires, enriching the dataset with subjective insights. A benchmark is established using essential features and standard machine learning methods to assess its utility.

3.6.3 DEAP Dataset

The DEAP dataset examines human emotional states and captures both neurophysiological and physiological signals. The dataset includes physiological signals and EEG recordings from 32 participants who viewed 40 one-minute-long segments of music videos. In order to improve the richness of the dataset, participants were asked to provide ratings for each video, assessing levels of dominance like/dislike, familiarity, valence, and arousal.

The dataset features frontal face recordings for 22 out of 32 subjects. This adds depth and context to facial expressions and provides an understanding of their link with the measured physiological signals while watching music videos.

The novel stimuli selection method is a combination of different approaches. The website includes video highlight detection, Last FM affective tags, and an online assessment tool to ensure diverse and emotionally engaging stimuli. This methodology aims to consider a great variety of affective responses using stimuli close to human life’s realities.

This multimodal dataset provides a pivotal resource to researchers investigating music photography’s human affective states under visual stimuli and music. The dataset available to researchers comprises, in addition to EEG data, peripheral physiological signals, subjective ratings, and facial expressions, thus adding to its usefulness in complex stimulus-physiology-subjective emotional experience interplay investigation.

3.6.4 AMIGOS Dataset

The presented database is a comprehensive resource for research on personality traits, affect, and mood, using signals related to neuro-physiology in a multimodal study. Unlike other databases, this collection elicits effects through long and short videos, integrating two configurations: individual groups of viewers and individuals. The primary objectives are to investigate individual affective responses about mood and personality and to analyze how these responses are influenced by the individual/group viewing configuration and video duration (short vs. long).

The databases were collected through two experiments involving 40 participants who watched short and long emotional videos alone or in groups. Neurophysiological signals, including ECG, GSR, and EEG, were captured using wearable sensors and full-body, frontal, and depth videos. Participants were also assessed for personality traits and mood using Positive and Negative Affect Schedules and the Big Five.

Emotions were annotated through external evaluation of arousal with valence and also self-assessment. The database supports analysis of correlations between these assessments and various factors, including video length, personality, mood, and social context. This allows for a detailed exploration of emotional responses in different scenarios.

3.6.5 DREAMER Dataset

DREAMER is a multimodal database for affect elicitation research featuring ECG and EEG signals recorded during audio-visual stimulus exposure from 23 participants. It includes self-assessment reports on valence, arousal, and dominance, using portable and cost-effective Emotiv EPOC and Shimmer2 sensors. The classification results for these affective states are comparable to those from databases using more expensive medical-grade equipment.

3.6.6 IEMOCAP Dataset

The IEMOCAP data release offers a rich dataset capturing affective interactions involving ten male and female actors. The database encompasses scripted and improvised sessions, totaling around 12 h of audio-visual content. Detailed information is provided for each recording, including audio and video of both participants, Motion Capture data of face, head, and hand movements for one interlocutor, text transcriptions of conversations, and word-level, phone-level, and syllable-level alignment. Additionally, annotations by multiple annotators offer categorical and dimensional labels for each utterance, enhancing the dataset’s depth for emotion-related research.

3.6.7 Affective Road Dataset

In driver monitoring, physiological measures of arousal and stress are developed and validated using two groups of wireless sensors. These sensors collect data on heart rate, electrodermal activity, respiration, skin temperature, and hand movements, while a separate network tracks environmental parameters like temperature, humidity, pressure, and light. This approach integrates physiological responses with road environment data to assess driver well-being and performance.

This multimodal dataset can be a valuable resource for researchers in this field of study, which involves music, visual stuff, and human affective. With the addition of EEG, peripheral physiological signals, subjective ratings, and facial expressions, we expanded the utility of this dataset beyond using it to investigate direct associations between stimuli and subjective emotional experiences.

3.6.8 The FER2013 (Facial Expression Recognition 2013) Dataset

Consists of seven emotions of grayscale images: Angry, Disgust, Sad, Surprise, Fear, Happy, and Neutral. The dataset includes 28,709 examples in the training set, 3589 images in the public testing set, and 3589 images in the private test set.

3.6.9 RAVDESS Dataset

The RAVDESS includes 7356 files (24.8 GB) from 24 professional actors (12 female, 12 male) speaking two lexically-matched statements with a North American accent. It features speech expressing seven emotions and song expressing five emotions, each at two intensity levels plus a neutral expression. The dataset is available in three formats: audio-only (.wav), audio-video (.mp4), and video-only.

3.6.10 SEED Dataset

The SEED dataset was designed to explore emotional responses to film clips. Seventy-two clips were chosen to elicit emotions such as happiness, sadness, fear, and neutrality. Fifteen subjects participated in three sessions, each viewing 24 clips, with EEG signals recorded using a 62-channel ESI conducted by NeuroScan System and eye movements tracked by SMI eye-tracking glasses.

The collected data were processed to extract critical features. EEG signals were down-sampled to 200 Hz and filtered to reduce noise, with Power Spectral Density (PSD) and Differential Entropy (DE) features extracted across the delta, theta, alpha gamma, and beta frequency bands. Additionally, eye movement features such as pupil diameter, fixation duration, saccades, and blink events were extracted from the recordings.

4  Use Cases by Symptoms and Disease

The use cases of FL mental health applications answer to a specific medical condition or a set of symptoms, as shown in Fig. 10. This method is flexible and enables individualized support or interventions to be offered. The following are several famous use cases characterized by symptoms and conditions.

Figure 10: Federated learning by mental health symptoms

4.1 Depression

Federated Learning Programs for Early Detection of Depression and Depression in Relation to Eating Disorders can incorporate behavioral analysis, the provision of individualized therapeutic material, and the capacity to monitor mood.

In 2021, Liu et al. enhanced the depression treatment robot (DTbot), an intelligent robot for depression therapy, using federated learning and conventional deep learning techniques. They solved the problems of central server-based models by localizing patient data, which optimizes treatment efficiency and addresses privacy issues. This is a potentially helpful integration of technology into mental health care [79].

Similarly, Xu et al. in 2021 addressed the problem of diagnosing depression and argued that the latter entails extremely inconsistent manifestations. The traditional centralized machine learning methods faced difficulty aggregating patient data, which must be highly confidential. The research adopted federated learning with a multi-view approach to depression analysis in cross-institutional to address privacy issues. The approach incorporated later fusion techniques to address inconsistent time series data. Comparative experiments demonstrated that with enough participants, the Federated Learning Framework achieves a prediction accuracy of 85.13%, a 15% improvement over local training. Even with fewer participants and sufficient data, the accuracy remained high at 84.32%, showing a 9% enhancement [80].

In 2021, Pranto et al. introduced a model combining deep neural networks and federated learning for monitoring mental health through social networking datasets, particularly keyboard input. Utilizing RNN and federated learning, the system detects daily depression levels. The global model is located at a central server, and users’ local devices test their keyboard data against it. Anonymized test results contribute to a global sentiment dictionary, which is updated daily. The model attained an accuracy of 93.46% on the 60th day, showing its effectiveness in continuously monitoring mental health through user-generated social media content [36].

In 2022, Bn et al. proposed FL for decentralized and privacy-preserving depression analysis in speech. Utilizing the DAIC-WOZ dataset, they demonstrated that federated models achieve a vital depression assessment with only a 4%–6% accuracy reduction compared to centralized approaches. These FL models perform better than previous methods, ensuring small memory footprints, energy efficiency, and short inference latency [48].

In 2022, Lemos et al. presented the Federated Artifcial Intelligence solution for moniToring mental Health status after cancer treatment (FAITH) study, which aims to remotely assess breast and lung cancer survivors in Spain and Portugal for signs of depression. They used markers such as nutrition, sleep, activity, and voice and employed standardized depression measures (Ham-D and HADS) for regular evaluations [81].

FAITH employs a smart band synced with a smartphone via the FAITH App to assess depression factors with minimal user input. The study aims to develop an FL Framework to predict depression while preserving data privacy. This model offers healthcare providers early alerts for timely mental health interventions, enhancing the quality of life for cancer survivors.

In 2022, Cui et al. examined the privacy concerns associated with centralized training depression diagnosis models based on speech. Utilizing FL, the study demonstrated for the first time that such models can be trained in a privacy-maintaining setting, maintaining decentralized speech data on users’ devices. The proposed FL model, integrated with defenses like norm bounding and differential privacy, proves robust against attacks. Experiments using the DAIC-WOZ dataset indicated that the FL model achieves great results comparable to centralized learning approaches, ensuring user privacy in speech data [47].

In 2022, Gupta et al. proposed an FL Framework using clustering to integrate multi-source data and improve standard machine learning models. It assesses the framework’s performance against other cooperative learning methods, highlighting its potential to maintain patient privacy across institutions [82].

In 2023, Lee et al. conducted across five South Korean tertiary hospitals, developed a federated, differentially private model to predict bipolar disorder transitions in patients initially diagnosed with depression. Utilizing data from 17,631 patients for model development and 4541 patients for external validation, the federated model outperformed local models with AUCs of 0.726 and 0.719, respectively. Key predictors included the Charlson index, severe depression, anxiolytics, young age, and visiting months, with bipolar transitions linked to seasonality, especially in spring and summer. This study indicated federated learning’s effectiveness in creating predictive models while preserving patient privacy [83].

In 2024, Khalil et al. explored FL for detecting depression in multilingual contexts from English, Arabic, Russian, Spanish, and Korean datasets. FL performed similarly to centralized and local models in IID scenarios but faced challenges in non-IID settings due to data disparities. Despite these challenges, FL holds promise as a privacy-protecting approach for mental health diagnostics [43].

In 2024, Liu addressed privacy concerns using social media data for depression detection by proposing a federated deep learning framework. This framework analyzed data from three social media platforms while preserving privacy and supporting clinical decision-making with accurate depression risk identification and intervention strategies. Experimental results indicated that the approach achieves state-of-the-art performance, offering a dynamic and effective clinical decision support system for personalized treatment [35].

In 2024, Gupta et al. focused on developing a privacy-preserving system for identifying Major Depressive Disorder (MDD) using EEG data through FL. FL allows for secure training of deep learning models without compromising patient privacy. The research evaluated various deep learning algorithms, including LSTM, GRU, Bi-LSTM, and 1D-CNN, for 3-channel EEG-based MDD screening using the MODMA dataset. Results showed that FL effectively trains models in a privacy-preserving way, with the potential for future research to expand and enhance MDD diagnosis methods [84].

Table 3 shows that most works on depression detection tasks implemented federated deep learning. The study highlights the detection of depression and its fine-grained level as well. The dataset came from several sources, including social networking data, mobile applications, standard datasets like DAIC-WOZ, and electronic health records.

4.2 Mental Stress

Federated learning applications designed for stress management utilize user data, including physiological signals, to create personalized interventions. These interventions may include real-time stress detection and adaptive coping strategies.

Another study relevant to this research is that of Vyas and Das, which deals with predicting driver stress and behavior for ADAS, which is essential in enhancing driving safety. The study estimates driver stress and driving behavior changes based on an analysis of historical trip data. This study uses LSTM with FCNN to predict driver stress levels. It also connects with driving behavior and stress and presents an intelligent recommendation system for taxi firms to propose appropriate drivers for future trips [85].

In addition, the proposed approach integrated FL in the Vehicular Edge Computing study. As a result, roadside units are integrated to do all data computations remotely. Testing the model on the UAH-DriveSet dataset demonstrated that this system can predict stress with 95% accuracy. The results showed that the model effectively enhances driving quality and the overall driving experience, suggesting that similar systems could improve road safety.

In 2021, Liu et al. evaluated the FL of the mHealth model, simulating affective state prediction using data from a chest-worn device. The FL model reached an accuracy of 92.8% in classifying neutral, stress-inducing, and amusement tasks, outperforming individual models that achieved 90.2%. These highlights federated learning’s potential in mHealth for secure, high-performance predictions [17].

In 2022, Fauzi et al. deployed an FL-based stress detection system and compared the individual, centralized, and FL methodologies using the WESAD dataset with LR as their classifier. This result showed that federated learning is less accurate when compared to both individual and centralized learning. This was followed by the average F1-measure for the individual learning paradigm, which was 99.96% and an accuracy of 99.98%. The lower accuracy shown in Federated Learning highlights the potential privacy-preservation vs. model performance trade-offs that will be of interest when applying stress detection systems into individual-data-privacy first scenarios [63].

In 2022, Rafi et al. monitored the Internal State of Drivers Towards Enhancing Riding Comfort for Car Occupants: Foundations and Development of Control Law for ADAS. The proposed driver’s stress monitoring system was based on physiological signals analysis to detect stressful conditions automatically. The system employed new-in-training about the individual differences among drivers [86].

The Federated Learning approach uses a personalization training regime that prioritizes samples from drivers with close posts like the target driver, hence customizing the prediction model according to every driver’s specific personality. This strategy is better than the conventional paradigms, which are optimized for average performance for all the users in each dataset. The AffectiveROAD public database study demonstrated the effectiveness of this personalization approach in improving stress estimation performance. Such results demonstrate the prospective of federated learning in driver stress monitoring, a more personalized and efficient approach in contrast to the traditional approaches [70].

A personalized, federated learning algorithm for stress-level classification was proposed by Jiang et al. In 2023. They demonstrated accuracy improvement by over 7% and 12%, respectively, compared to two recently introduced baseline algorithms under the proposed approach. Also, the two baseline algorithms realized a significant decrease of 37.5% and 9.6% in training time. They introduced a novel cold-start algorithm designed for integrating new clients into the trained system, achieving robust performance in individual classification accuracy and overall training time [87].

Federated Deep Learning Based Wearable Electrodermal Activity for Preserving Chronic Aging Patient Data Privacy. The research utilizes the WESAD dataset, encompassing five data categories: distraction, distress, and meditation experience. Pre-processing transforms the raw dataset with SMOTE and min-max normalization. The FL-based approach trains their DNN model independently on the dataset when two users forward updates to the model. Each client analyses the results three times to overcome the problem of overfitting [88].

Experimental results indicated that when the FL-based DNN model is used, it delivers 86.82% accuracy and ensures patient data privacy. This approach outperforms prior studies in detection accuracy, underscoring the potential of FL-based DNN models for enhancing advanced stress classification from wearable data while preserving vital patient data privacy.

In 2021, Can et al. used FL to analyze the heart activity data collected from the smart bands and monitor the stress levels during different events. The federated learning application showed promising results exhibiting the relevance of wearable IoT-based biomedical monitoring devices. The method maintained the secrecy of the extracted information, demonstrating the efficiency of federated learning in this domain [89].

In 2023, Vyas et al. developed FedSafe as a driver stress-oriented recommendation system using FL to improve driving experience and wider transportation system benefits. FedSafe uses vehicle telematics and physiological data to explore stressed conditions and driving habits, providing personalized advice for the next trip to minimize on-road stress and secure trips. Collaborative learning using vehicle data without transferring raw information among drivers, thus increasing transmission data overhead, is made possible through incorporating Federated Learning. Such a method is also appropriate for the vehicular network whose popularity requires edge computation [90].

An extensive human-in-the-loop study using publicly available natural driving datasets (UAH DriveSet, HCI Lab, PhysioNet) demonstrated that the proposed FedSafe system shows impressive predictive capabilities. It achieved 97% accuracy in predicting driver stress and 98% in identifying driving behavior, measured by AUC and F-measure. The Federated Learning approach substantially reduces transmission overhead by approximately 25 times, indicating the efficiency and effectiveness of FedSafe in stress monitoring and behavior recognition for enhanced road safety and driving experiences.

In 2024, Alahmadi et al. explored using the Internet of Medical Things for remote patient monitoring and data-driven diagnosis. They presented use cases and recent developments in FL-based IoMT data analysis, particularly in mental stress detection. The proposed framework demonstrated that FL can significantly reduce communication overhead for local devices, from 10.02 MB/day to as little as 754 B/day, compared to traditional non-FL methods. This reduction enhances the efficiency and practicality of using FL in IoMT for continuous health monitoring (39).

Table 4 summarizes all the federated learning in stress identification tasks, indicating that federated deep learning is prevalent in these tasks. In addition, the dataset involves driver stress data sets such as the AffectiveROAD dataset and the UAH-DriveSet dataset. The standard dataset, such as the WESAD dataset, collected the signals from chest wear and wristbands and integrated them into these tasks. The dataset was gathered from several sources, including social networking data, mobile applications, standard datasets like DAIC-WOZ, and electronic health records.

4.3 Sleep Behavior

Federated learning contributes to applications targeting sleep-related issues by analyzing data from wearables and devices. These apps provide insights into sleep patterns, identify factors affecting sleep quality, and offer personalized recommendations for improved sleep hygiene.

In 2022, Khao et al. applied ML to analyze and predict sleep quality. They took things further by introducing FL into the analysis. The research used an ML model, FedMCRNN, incorporating multi-modal data collected from wearable sensors of sleep quality prediction. In the many-to-one setting, accuracy was 96.774%. In the many-to-many setting, accuracy was 68.721%. Federated learning using multiple convolutional recurrent neural networks (FedMCRNN) outperforms traditional ML models and prevails over other advanced methods for sleep quality prediction. This is important for sleep analysis, and FL can also support doctors using AI, demonstrating that it can increase predictive accuracy and feature understanding [91].

In 2022, Yin et al. investigated EEG data for sleep staging and proposed FL as a solution. They employed the K-means partitioning algorithm to convert EEG datasets into non-independent and identically distributed unbalanced datasets simulating various user datasets. MLP networks were used for the analysis, and the research examined privacy leakage problems in Florida. The user’s EEG data was protected using DP. The research indicated that errors are related to the disparity in client-based data distribution between L and CL [51].

In 2023, Anido-Alonso et al. proposed several strategies for deep learning networks to achieve inter-database generalization while ensuring patient data privacy during automatic sleep staging. Implementing the sub-sampled federated stochastic gradient descent (ssFedSGD) algorithm with four ensemble strategies, the decentralized learning approaches outperformed baseline local methods and matched the generalization results of centralized approaches. The findings showed that decentralized methods provide advantages in flexibility, scalability, and data privacy [92].

In 2024, Ma et al. addressed the challenge of limited labeled data in automatic sleep staging by proposing an FL strategy for secure collaboration among institutions. The Federated Semi-Supervised Sleep Staging method enhances labeled data and task knowledge by integrating external knowledge from other institutions using single-channel EEG data. It incorporates local and global relationships, employs pseudo-labeling optimization for reliable pseudo-labels, and uses prototype-contrastive learning to clarify relationships. Experimental results showed that this method outperforms comparative approaches on two publicly available datasets [93].

In 2024, Borges et al. explored FL for sleep detection in human-in-the-loop Cyber-Physical Systems, addressing privacy concerns with sensitive user data. FL achieved over 84% accuracy, comparable to traditional machine learning models. The research emphasized the importance of managing human error inputs, demonstrating FL’s potential as a privacy-preserving and accurate solution for sleep detection [94].

In 2024, Dautov et al. studied developing an FL system for drowsiness detection using non-IID data. They compared the performance of this federated setup with a traditional centralized approach. The results identified that the centralized model achieved approximately 83% accuracy and recall precision, while the federated model’s best performance reached around 66%, with an average performance of about 50% [95].

Table 5 presents previous literature on federated learning in sleep behavior. It indicates that federated deep learning is often and primarily employed to monitor sleep behavior using EEG data, a brain signal.

4.4 Emotion Detection (Most of the Work Presented)

Federated learning allows applications to detect mood variations, identify triggers, and provide timely solutions for people suffering from mood disorders [96]. These interventions can be anything from activities that improve mood to putting consumers in touch with mental health specialists.

In 2020, Chhikara et al. integrated speech signals and facial expressions to identify micro expressions and generate an emotion feature, providing insights into users’ mental health. Utilizing federated learning, the proposed scheme allows clients to train the model locally without breaking privacy. Instead of sending raw data to a central server, only model weights are transmitted, creating a more robust global model that is sent back to users. This approach ensures inter-organizational training without violating privacy, leading to good results. The collected user data is monitored to analyze mental health, offering counseling solutions during challenging times [56].

In 2020, Latif et al. applied SER applications with federated learning, which allows collaborative learning without sharing raw speech data. The study achieved good results compared to novel methods by evaluating this approach using the IEMOCAP dataset. The dataset included four basic emotions (angry, neutral, happy, and sad) from the IEMOCAP corpus, and the proposed system employed CNN and RNN architectures for emotion classification. The LSTM-based classifier slightly outperforms CNNs [69].

In 2022, Nandi et al. assessed the interconnectedness of emotional and physical health, emphasizing the significance of simultaneous care for both aspects. It explored affective computing, focusing on detecting emotional states through physiological measurements such as electrodermal activity data (EDA), respiratory belts (RB), and EEG. Focusing on real-time multi-modal emotion state recognition from wearable sensors emerges as a prominent topic. The challenge lies in processing high data generation rates and ensuring data privacy, particularly in decentralized settings [65].

In order to address these challenges, the study introduces Fed-ReMECS, a federated learning framework established for the real-time detection of emotion states within multimodal physiological data streams obtained from wearable sensors. This structure works well and allows it to construct a global classifier without touching users’ local data, causing privacy issues in the IoT environment. The proposed algorithm, Fed-ReMECS, can protect users’ privacy while maintaining efficiency and scalability, as demonstrated through experimental studies conducted with the DEAP dataset. The results show that it can be employed to develop robust and privacy-protecting emotion classification systems for real-time usage in IoT [65].

In 2022, Singh et al. introduced an FL Approach designed to classify emotions into seven categories: happiness, excitement, anger, empathy, horror, nervousness, and astonishment. The priority is privacy when it comes to the proposed framework because, although it can classify data with accuracy like the conventional centralized machine learning solution, it does it in a confidential manner whereby individual data is not exposed [97].

In 2022, Xu et al. proposed an extension of Federated Learning to address the problem of scarce data available for training emotion detection models based on electroencephalogram signals. With FL, several clients’ cooperation is possible without disclosing information and, thus, privacy concerns. The proposed framework achieved a high accuracy of 90.74% in emotion recognition on the SEED and DEAP datasets. When applied to DEAP dataset clients, the FL method outperformed local training by 29.31%, emphasizing the necessity of FL for emotion recognition tasks. In addition, the study highlighted non-IID data on FL training, demonstrating a 14.89% decrease in accuracy compared to IID data [52].

In 2022, Salman et al. implemented FER on small smart devices due to computational constraints. They proposed a real-time image- and video-based facial expression recognition model that overcomes these limitations. The approach involved personalization through fine-tuning a central model with unlabelled videos from the target subject using an FL strategy. The image facial expression recognition model generated pseudo-labels, and adaptation was performed by averaging local model weights. Results demonstrated that this method improves edge device and central video facial expression recognition model performance, achieving a 13.1% gain in F1-scores for personalized models and up to a 3.4% improvement in the central video facial expression recognition model’s mean micro F1-score. The lightweight solution is well-suited for deployment on small smart devices [57].

In 2022, Zhang et al. implemented facial expression cognition technology, emphasizing the privacy concerns in current distributed machine learning for face monitoring data. The proposed solution introduces an FL framework for unsupervised face recognition post-exercise. Employing a graph AE design, the framework enables collaboration among multiple edge devices to optimize a common objective function, enhancing global model efficiency. A multidomain learning loss function promotes shared feature representation across tasks, bolstering domain adaptability. Adversarial learning further improves recognition within each domain. Experimental validation on diverse multidomain expression datasets demonstrated a 19% increase in F1-score compared to the benchmark approach [98].

In 2023, Chen proposed an online teaching emotion recognition system with federated learning for assessing AI-assisted teachers’ teaching effectiveness. Although the current focus in intelligent education centers on enhancing students’ learning efficiency, this proposal emphasizes the importance of evaluating teachers’ educational impact. The Federated Learning approach ensures the security, privacy, and legal compliance of teachers’ data while enabling collaborative training and joint modeling. By recognizing teachers’ teaching emotions, the system aids in analyzing teaching levels, identifying issues in online teaching processes, and enhancing overall teaching proficiency. Experimental results demonstrate the model’s effectiveness in recognizing facial emotions during online education, especially in scenarios with independent and unbalanced data among participants [99].

In 2023, Agarwal et al. raised privacy concerns in emotion recognition from physiological signals by proposing a Federated Learning Framework. Unlike centralized datasets, this decentralized approach preserves privacy, essential for sensitive data such as EEG recordings. The convolutional neural network model directly processed EEG signals to extract features from the DEAP dataset. The proposed architecture achieved accuracy scores of 72.22%, 70.10%, and 66.99% for dominance, arousal, and valence, respectively, by maintaining the non-IID nature of the data. The study discussed trade-offs between performance and privacy in Federated Learning [66].

In 2023, Zhang et al. introduced FedAudio, designed to address the lack of audio data and tasks in existing FL benchmarks. With the increasing concern for data privacy on consumer devices, FedAudio aims to facilitate research in Florida for audio-related tasks. The benchmark includes four representative audio datasets, spanning three essential audio tasks relevant to FL use cases. A specific feature of FedAudio is the incorporation of data noises and label errors into the datasets, simulating challenges encountered in real-world FL system deployments. The benchmark provides results that enable researchers to compare algorithms effectively. FedAudio is anticipated to inspire and advance FL research in audio tasks, benefiting the speech research community. Access to datasets and benchmark results is available [50].

In 2024, Gahlan et al. introduced a Federated Learning (FL) approach based on the Multi-modal Emotion Recognition System (F-MERS) designed for assessing complex emotions using EEG, ECG, GSR, and respiratory therapy (RESP) physiological sensor data. In contrast to traditional machine learning classifiers that risk privacy violations, F-MERS employs FL, allowing local model updates without exposing raw training data. Utilizing MLP as the base model, F-MERS achieved accuracies of 87.90%, 89.02%, and 79.02% on the DEAP, AMIGOS, and DREAMER datasets, respectively. The study explored scalability, efficient communication, and privacy preservation compared to a non-FL MLP baseline model. With multi-modal physiological sensors, this represents the first FL-enabled framework for detecting complex emotions in three dimensions (valence, arousal, and dominance) [40].

Table 6 illustrates that most emotion tasks using federated deep learning and emotion datasets have various forms, such as speech and facial expression. Combining both data types helps to make more robust models for emotion recognition. DEAP datasets containing EEG, peripheral physiological, and audio-visual recordings are primarily employed in federated learning tasks.

4.5 Other Applications

Other mental health applications use federated learning to address mental health concerns such as loneliness detection and mental health symptom detection.

In 2022, Qirtas et al. highlighted loneliness among college students needing improved technology-based detection methods considering individual behavioral differences. The research proposed loneliness detection, utilizing passive sensing data from mobile phones to capture rich user behavioral information. Employing unsupervised clustering, the study identified small groups of students with similar behavioral patterns, allowing for the understanding and detection of loneliness. By creating group-specific classification models, the research demonstrates an improvement in accuracy over generalized models. The findings underscored personalized approaches in mental health interventions and highlighted the value of passive sensing data for creating tailored loneliness detection approaches [100].

In 2022, Ahmed et al. addressed the need for diverse patient data in training DNNs for mental health symptom detection, introducing a structural hypergraph with an emotional lexicon and a federated learning-based embedding model. The model used consecutive word sequences and attention-based mechanisms to enhance semantic word representations, incorporating grammatical, vocabulary, and dynamic lexicon analysis. Experimental results showed that using a Bi-LSTM architecture with an attention mechanism, the approach achieved a region of convergence (ROC) of 0.86, demonstrating its effectiveness in encoding emotional words and improving clinical applicability [101].

In 2024, Sahu et al. explored federated learning for private and accurate AI-based schizophrenia diagnoses using distributed datasets while preserving individual privacy. Advanced AI algorithms were applied to diagnose schizophrenia, demonstrating federated learning’s potential as a secure and effective method for mental health diagnostics. The study highlighted federated learning’s ability to enhance diagnostics while maintaining data confidentiality [102].

In 2024, Aversa et al. explored using FL to enhance technologies for monitoring, diagnosing, and managing specific health conditions while prioritizing privacy, confidentiality, and cybersecurity. The project aims to address the current lack of privacy-respecting solutions in the market. REDRAW focused on studying and refining dynamic cloud-edge deployment techniques that utilize FL models across three real-world contexts. The goal was to improve existing technological solutions while adhering to the strategic and non-functional constraints prevalent in the Italian and European healthcare landscapes [103].

With a focus on other mental health applications, the work also highlighted areas such as wellness detection, loneliness detection, and other mental disorders like schizophrenia. The methods employed range from federated unsupervised learning to federated deep learning. The primary concern with using federated learning in this context was to protect the privacy of mental health data.

5  Backgrounds for Federated Learning

The main techniques used in the research regarding federated mental health learning are Federated Machine Learning and Federated Deep Learning.

5.1 Federated Machine Learning

FL revolutionizes the conventional model training landscape by decentralizing the process across multiple devices or servers, each holding local data samples. The primary reason is to safeguard privacy and security, especially where sensitive data such as financial and personal details are concerned. This approach allows collaborative training over various devices without exposing raw data—a giant step towards protecting user privacy [104–106].

Federated learning is one of the models with its core decentralized training. Instead of relying on a centralized server, FL implements the training phase on individual devices or servers in a decentralized manner. The data being raw is localized to maintain privacy as the first objective. Information disclosure risks are reduced as only model updates or aggregated information is shared [107,108].

One critical factor is collaborative learning, where several people participate and improve the model using the data specific to each participant, which enhances collaboration and increases knowledge among the participants [85]; a machine learning model for global training. Consolidating each participant’s update develops a global model with aggregated knowledge that does not disclose specific individual data. The adaptability of federated learning for heterogeneous environments and its efficient communication and collaborative structure positions it as a potent paradigm for emotion recognition [109].

Fig. 11 focuses on the association between the local devices and the global server. There is an in-depth explanation of how this exchange is conducted as follows:

•   Local Model Training

Each local device or server possesses its dataset, typically representing a user’s data or a subset of users. The local model on each device is initially trained on its respective dataset. This training process utilizes the local data to update the model’s parameters [110].

•   Model Updates

Rather than transmitting raw data to a central server, which could break user privacy, only the updates to the local model are transmitted. The local device calculates the variance between the existing and refined model parameters post-training. These updates, often as gradients or weights, encapsulate the knowledge gained during local [111].

•   Secure Transmission

These model updates are then transferred to the global server securely. Common cryptographic techniques, such as encryption, are employed to ensure the confidentiality of the exchanged information. Using secure transmission protocols prevents unauthorized access to the model updates during their process from local devices to the global server [104,112,113].

•   Aggregation at the Global Server

Upon receiving updates from multiple local devices, the global server combines these updates to create a comprehensive global model. Aggregation methods vary but commonly involve averaging the model parameters across devices. This process captures the collective knowledge from all participating devices without exposing individual data [114].

•   Global Model Distribution

The combined global model is then sent back to the local device. This generates a model that contains these globally applicable insights. In this regard, the local devices synchronize their models with the received global model, so every device enjoys collaborative knowledge without raw data exchange. Compared to the centralized model training that involves sending out sensitive information via many devices, Federated learning reduces the amount of data transmitted as it does not send all information on the models but sends model updates, which are later summed up at the global server, hence protecting the privacy [107].

Figure 11: The link between local nodes and a central server

In the federated learning process, each of the K clients possesses its local dataset (Dk) without sharing it externally. The objective is to minimize a global loss function L(w) in a distributed manner. This global loss function represents the model’s overall performance across all clients. The loss function (Lk) for the k-th client is defined as the average loss over its local dataset (Dk) in Eq. (1):

Lk=1nk∑i∈Dkli(W)(1)

where w is the global model parameters, li(W) is the loss function for a specific data sample i in the local dataset Dk, and nk is the size of the local dataset for the k-th client.

The global goal of federated learning is to minimize the weighted average of these local client loss functions as Eq. (2), where the weights are proportional to the size of each client’s dataset (nk). The weighted average is calculated over all K clients in Eq. (2):

minL(W)=min∑k=1knknLk(W)(2)

The total data sample size for all clients = n. The optimization process, in which all the clients take part in minimizing the global loss function, highlights the collaborative essence of federated learning. This approach ensures that the model is trained using data sets of different types without disclosing information about personal clients. Algorithms 1, 2 depict a federated learning process, highlighting its distributed and collaborative characteristics [115].

5.2 Federated Deep Learning

Federated deep learning is a class of federated learning techniques that incorporates the features of deep neural networks. This is especially important in deep learning, where complex models are usually used. Unlike centralizing the training process on one server, federated deep learning distributes it across several servers and/or devices, each with its sample of locally available data. One of its primary objectives, aligning with Federated Learning principles, is to address privacy concerns. It achieves this by ensuring raw data is on local devices, with aggregated information or only model updates being shared to prevent the exposure of sensitive information [116,117].

The collaborative nature of federated deep learning involves participants contributing to global deep neural network training. Each participant refines the model using its local dataset, and these refinements are aggregated to construct an enhanced global model. Efficiency in communication is a hallmark of federated deep learning, focusing on minimizing the exchange of information between devices [118,119]. This approach is valuable in scenarios where communication resources are limited. Federated deep learning is designed to adapt seamlessly to heterogeneous environments, accommodating devices with differing data distributions. It scales effectively for large models, which are common in tasks, for example, speech recognition, image recognition, or natural language processing [120–123]. Federated deep learning aspires to create robust and generalized models that perform effectively across various scenarios by learning from diverse datasets across participants [124–126].

5.3 Software Packages for Federated Learning

A combination of general-purpose ML and DL libraries and specialized federated learning frameworks is often used for Federated Learning. The following is an overview of the software packages and tools commonly involved:

5.3.1 Machine Learning and Deep Learning Models

- Scikit-learn: Used for traditional machine learning models and evaluation metrics. It is commonly used for tasks such as preprocessing, model selection, and performance evaluation in federated learning [40,101].

- Karas: A high-level neural networks application programming interface (API), often used with TensorFlow as a backend for building and training deep learning models [65].

- TensorFlow: A comprehensive open-source library for machine learning and deep learning. TensorFlow Federated (TFF) is an extension for implementing Federated Learning using TensorFlow models [48,65,94,101].

5.3.2 Federated Learning Frameworks

- TensorFlow Federated: An open-source framework for federated learning that trains machine learning models across decentralized data sources while ensuring privacy. It is integrated with TensorFlow to use deep learning models in federated settings [22,40].

- Flower: [127] A flexible framework for federated learning that supports different machine learning and deep learning libraries, including TensorFlow and PyTorch. It is designed to be easy to use and integrates well with existing machine learning workflows [43,83,84,94,100].

- PySyft: [128] developed by OpenMined, is a Python library designed to facilitate the implementation of training algorithms with privacy mechanisms. It supports PyTorch and TensorFlow and offers features like secure multi-party computation and differential privacy (22).

5.3.3 Voice-Related Tasks

- Librosa: A Python package for analyzing and processing audio signals. It is often employed for extracting features like Mel-frequency cepstral coefficients (MFCCs), used in voice-related federated learning tasks [56].

5.3.4 Text-Related Tasks

- Hugging Face Transformers: A library for natural language processing tasks that provides pre-trained models for text classification, sentiment analysis, and text generation. It is often used in federated learning for text-related tasks [43].

5.3.5 Image-Related Tasks

- OpenCV: A library of programming functions mainly aimed at real-time computer vision. It is commonly used for image preprocessing in federated learning [129].

These packages allow federated learning across various data modalities, ensuring that machine learning and deep learning models can be applied to a range of tasks while leveraging the benefits of federated learning.

5.4 Performance Metrics of Federated Machine Learning and Deep Learning

Accuracy Metrics

Accuracy measures the proportion of correctly predicted instances, while precision and recall provide insights into positive predictions and identify relevant instances. The F1-score balances precision and recall, and the AUC-ROC curve evaluates the model’s ability to distinguish between classes, reflecting its overall discriminative power [130,131].

Loss Metrics

Regarding federated learning, loss functions are critical in evaluating how well a model’s predictions align with the actual values. Common loss functions include Cross-Entropy Loss for classification tasks and Mean Squared Error for regression tasks. In a federated context, calculating the average loss across all participating clients can provide a clear picture of the model’s performance and effectiveness in learning from distributed data sources [132].

Communication Efficiency

Communication efficiency is crucial in federated learning, impacting model training speed and practicality. It involves managing communication overhead, the data exchanged between clients and the server, communication costs, and the time and resources required for data transfer. Techniques such as model compression and efficient aggregation methods help minimize these costs, improving overall system efficiency [133].

6  The Process for Federated Learning in Mental Health

The steps required to set up a federated learning system between a central server and a mental health facility (MH facility) are outlined in the sequence diagram in Fig. 12. The MH Facility requests the Federated Learning Setup from the Central Server after identifying the problem and process objectives. The MH Facility’s desired model architecture is then described.

Figure 12: The sequence diagram for establishing federated learning in mental health application

The central server verifies the architecture needed for the Federated Learning Process. The MH Facility then prepares the data, separating it into training and validation datasets before forwarding the anonymized, pre-processed data to the central server. The central server develops a global model, and the MH Facility modifies a local model in the Federated Learning Architecture. Secure lines are available for communication between the MH Facility and the central server.

The Central Server requests a model review, and the MH Facility employs privacy preservation strategies. Iteratively, the central server asks for more training depending on the MH Facility’s evaluation results. The revised model is subsequently transmitted back to the central server.

After a successful development phase, the global model is distributed from the central server to the MH Facility. Ethics and compliance are ensured throughout the process, and the MH Facility acts to address bias and injustice. The Federated Learning Architecture is updated as needed to address difficulties in healthcare while upholding ethical principles and privacy.

7  Benefits of Federated Learning in the Mental Health Field

FL emerges as a transformative approach in the mental health domain, offering a range of pivotal advantages, as shown in Fig. 13. Regarding privacy issues, FL’s main advantage is that this allows the users’ sensitive mental health data to stay in separate devices and not be aggregated into a single place. This case refers to sensitive data, and thus, it has crucial privacy issues. FL enhances data security by minimizing the chances of a data breach that could lead to undesirable results such as discrimination, stigmatization, and other mental health issues. Further, FL offers collaborative learning by training models and several peers’ thoughts but shields individual information. Thus, a participatory approach is significant for mental health research [134–136].

Figure 13: Benefits of federated learning

In addition, FL supports heterogeneity for mental health data, allowing model training with diverse inputs that are adaptive to multiple mental health situations among populations and demographics. Routine screening of mental health indicators in FL for edge devices and wearables supports dynamic and responsive mental health analysis. Further, FL reduces communication inefficiencies by only sharing models’ updates or aggregated data in costly or low bandwidth environments [137–139].

FL is ethically sound as it aligns with notions of user autonomy that allow individuals to participate in mental health research and own their data. It can support a few research teams undertaking large-scale mental health research worldwide and allow cross-research teams to analyze trends and patterns collectively [140]. The continuous refinement of models based on insights from different participants underscores FL’s adaptability in the ever-evolving mental health field. Overall, FL significantly benefits mental health applications, paving the way for more effective, collaborative, and privacy-preserving approaches to mental health research and analysis [141].

8  Future Challenge Using Federated Mental Health Learning

Although FL holds great results in advancing mental health research and applications, several challenges and considerations must be focused on to unlock its full potential. The future challenges related to federated learning in mental health are illustrated in Fig. 14.

Figure 14: The future challenge of federated mental health learning

8.1 Heterogeneity of Data

Mental health data can vary widely regarding types of disorders, severity, and individual experiences. Handling this heterogeneity in Federated settings is challenging, and developing models that generalize well across diverse mental health conditions remains a key concern [142].

8.2 Data Imbalance

Some mental health conditions can be more prevalent than others, leading to imbalanced datasets. Ensuring fair representation of all conditions in federated learning models is crucial for accurate and unbiased predictions [143].

8.3 Model Interpretability

For instance, the federated learning schemes involving deep neural networks could be confusing. Successful federated learning adoption mandates creating models all involved mental health specialists know and accept as reliable [144–146].

8.4 Security and Trust

The use of federated learning involves several entities working together. Security of models’ updates, poisoning attacks mitigation, and trust between stakeholders are still existing hurdles for Federated mental health learning [147–149].

8.5 Dynamic Nature of Mental Health

Mental health conditions are dynamic. Creating federated models responsive to transient changes in mental health states and treatment responses necessitates perpetual assessment and model alteration [150–152].

8.6 Data Quality and Reliability

As far as data from different devices is concerned, it is necessary for federated learning to ensure that its quality and reliability are appropriately maintained. It is critical for the FL’s effectiveness in terms of mental health to address issues such as noise, outliers, and anomalies [153].

8.7 Ethical Considerations

Federated learning in mental health raises ethical considerations related to consent, transparency, and the responsible use of data. Balancing the benefits of collaborative research with protecting individual privacy and rights is an ongoing challenge [154].

8.8 Regulatory Compliance

Stringent privacy regulations frequently govern mental health data. Federated Learning must conduct while ensuring compliance with Health Insurance Portability and Accountability Act standards. HIPAA in the United States presents significant challenges that require careful attention [155].

8.9 Integration with Clinical Workflows

Integrating federated learning models into existing clinical workflows and electronic health record systems is crucial for their real-world applicability. Overcoming technical and organizational barriers for smooth integration remains challenging [156].

8.10 Longitudinal Studies

Federated learning models may benefit from longitudinal data to capture the progression of mental health conditions over time. Designing and conducting federated longitudinal studies pose logistical and methodological challenges [157].

Addressing these challenges will require interdisciplinary collaboration involving researchers, mental health professionals, data scientists, and policymakers. Overcoming these obstacles will contribute to the responsible and effective use of federated learning in advancing mental health research and applications.

9  Conclusions

This extensive study underscores the growing connection between technology and, more specifically, federated learning in mental health applications to combat the global burden of mental performance. Many applications arise from the growing need for mental health support, but privacy concerns regarding sensitive mental health data have been a significant hindrance.

In this context, this article investigates different federated learning models related to mental health topics. The advantages of federated learning in mental health applications are highlighted in this research. As the study looks forward, it proposes additional work to improve and modulate federated learning techniques especially for mental health purposes. This could, for example, entail the investigation of trainable predictive models via more advanced federated learning frameworks or adaptation of algorithms to mental health data or even, evaluating implications pertaining to continuity in utilizing federated learning for support on mental health. This research adds to the continually shifting mental health apps terrain by providing insight into status and direction for future work, highlighting federated learning as a crucial entity in maintaining privacy, safety, and cooperation in the world of digital mental health care.

However, one limitation of this study is that the applications of federated learning in mental healthcare are still in their early stages, with limited real-world implementations and long-term evaluations. Security and privacy concerns remain critical, as federated learning models are not entirely immune to data leakage or adversarial attacks.

Acknowledgement: Princess Nourah bint Abdulrahman University Researchers Supporting Project number (PNURSP2024R 343), Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia.

Funding Statement: Princess Nourah bint Abdulrahman University Researchers Supporting Project number (PNURSP2024R 343), Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia.

Author Contributions: Methodology, visualization, and writing—original draft: Vajratiya Vajrobol, Brij B. Gupta; conceptualization, visualization, validation, and writing—review and editing: Geetika Jain Saxena, Razaz Waheeb Attar, writing—review, and editing: Sanjeev Singh, Akshat Gaurav, validation, and writing—review and editing: Amit Pundir, Mosiur Rahman, Savi Bansal. All authors reviewed the results and approved the final version of the manuscript.

Availability of Data and Materials: All data generated or analyzed during this study are included in this published article.

Ethics Approval: This article does not contain any studies with human participants or animals performed by authors.

Conflicts of Interest: The authors declare no conflicts of interest to report regarding the present study.

References

1. Harerimana B, Forchuk C, O’Regan T. The use of technology for mental healthcare delivery among older adults with depressive symptoms: a systematic literature review. Int J Ment Health Nurs. 2019;28(3):657–70. doi:10.1111/inm.12571. [Google Scholar] [PubMed] [CrossRef]

2. Bucci S, Schwannauer M, Berry N. The digital revolution and its impact on mental health care. Psychol and Psychother Theory, Res and Pract. 2019;92(2):277–97. doi:10.1111/papt.12222. [Google Scholar] [PubMed] [CrossRef]

3. Banabilah S, Aloqaily M, Alsayed E, Malik N, Jararweh Y. Federated learning review: fundamentals, enabling technologies, and future applications. Inf Process Manag. 2022;59(6):103061. doi:10.1016/j.ipm.2022.103061. [Google Scholar] [CrossRef]

4. Rahman A, Hossain MS, Muhammad G, Kundu D, Debnath T, Rahman M, et al. Federated learning-based AI approaches in smart healthcare: concepts, taxonomies, challenges and open issues. Cluster Comput. 2023;26(4):2271–311. doi:10.1007/s10586-022-03658-4. [Google Scholar] [PubMed] [CrossRef]

5. Sheller MJ, Edwards B, Reina GA, Martin J, Pati S, Kotrotsou A, et al. Federated learning in medicine: facilitating multi-institutional collaborations without sharing patient data. Sci Rep. 2020;10(1):12598. doi:10.1038/s41598-020-69250-1. [Google Scholar] [PubMed] [CrossRef]

6. Xu J, Glicksberg BS, Su C, Walker P, Bian J, Wang F. Federated learning for healthcare informatics. J Healthc Inform Res. 2021;5:1–19. [Google Scholar] [PubMed]

7. Patel VA, Bhattacharya P, Tanwar S, Gupta R, Sharma G, Bokoro PN, et al. Adoption of federated learning for healthcare informatics: emerging applications and future directions. IEEE Access. 2022;10:90792–826. doi:10.1109/ACCESS.2022.3201876. [Google Scholar] [CrossRef]

8. Ali M, Naeem F, Tariq M, Kaddoum G. Federated learning for privacy preservation in smart healthcare systems: a comprehensive survey. IEEE J Biomed Health Inform. 2022;27(2):778–89. [Google Scholar]

9. Joshi M, Pal A, Sankarasubbu M. Federated learning for healthcare domain-pipeline, applications and challenges. ACM Trans Comput Healthc. 2022;3(4):1–36. doi:10.1145/3533708. [Google Scholar] [CrossRef]

10. Dasaradharami Reddy K, Gadekallu TR. A comprehensive survey on federated learning techniques for healthcare informatics. Comput Intell Neurosci. 2023;2023(1):8393990. doi:10.1155/cone.v2023.1. [Google Scholar] [CrossRef]

11. Li H, Li C, Wang J, Yang A, Ma Z, Zhang Z, et al. Review on security of federated learning and its application in healthcare. Future Gener Comput Syst. 2023;144:271–90. doi:10.1016/j.future.2023.02.021. [Google Scholar] [CrossRef]

12. Gahlan N, Sethia D. Federated learning in emotion recognition systems based on physiological signals for privacy preservation: a review. Multimed Tools Appl. 2024;1–69. [Google Scholar]

13. Khalil SS, Tawfik NS, Spruit M. Exploring the potential of federated learning in mental health research: a systematic literature review. Appl Intell. 2024;54(2):1619–36. doi:10.1007/s10489-023-05095-1. [Google Scholar] [CrossRef]

14. Kallakuri S, Gara S, Godi M, Yatirajula SK, Paslawar S, Daniel M, et al. Learnings from implementation of technology-enabled mental health interventions in India: implementation report. JMIR Med Inform. 2024;12:e47504. doi:10.2196/47504. [Google Scholar] [PubMed] [CrossRef]

15. Chaudhry BM, Debi HR. User perceptions and experiences of an AI-driven conversational agent for mental health support. Mhealth. 2024 Jul 5;10 : 22. doi:10.21037/mhealth-23-55. [Google Scholar] [PubMed] [CrossRef]

16. Wang T, Du Y, Gong Y, Choo KKR, Guo Y. Applications of federated learning in mobile health: scoping review. J Med Internet Res. 2023;25:e43006. doi:10.2196/43006. [Google Scholar] [PubMed] [CrossRef]

17. Liu JC, Goetz J, Sen S, Tewari A. Learning from others without sacrificing privacy: simulation comparing centralized and federated machine learning on mobile health data. JMIR Mhealth Uhealth. 2021;9(3):e23728. doi:10.2196/23728. [Google Scholar] [PubMed] [CrossRef]

18. Bhatt P, Liu J, Gong Y, Wang J, Guo Y. Emerging artificial intelligence-empowered mhealth: scoping review. JMIR Mhealth Uhealth. 2022;10(6):e35053. doi:10.2196/35053. [Google Scholar] [PubMed] [CrossRef]

19. Yu H, Reiff-Marganiec S. Learning disease causality knowledge from the web of health data. Int J Semant Web Inf Syst. 2022;18(1). doi:10.4018/IJSWIS.297145. [Google Scholar] [CrossRef]

20. Li J, Jiang M, Qin Y, Zhang R, Ling SH. Intelligent depression detection with asynchronous federated optimization. Complex Intell Syst. 2023;9(1):115–31. doi:10.1007/s40747-022-00729-2. [Google Scholar] [PubMed] [CrossRef]

21. Li J, Zhang R, Cen M, Wang X, Jiang M. Depression detection using asynchronous federated optimization. In: 2021 IEEE 20th International Conference on Trust, Security and Privacy in Computing and Communications (TrustCom), 2021; Shenyang, China; p. 758–65. doi:10.1109/TrustCom53373.2021.00110. [Google Scholar] [CrossRef]

22. Tabassum N, Ahmed M, Shorna NJ, Sowad UR, Mejbah MD, Haque HM. Depression detection through smartphone sensing: a federated learning approach. Int J Interact Mobile Technol. 2023;17(1):40–56. [Google Scholar]

23. Xiao J, Liu X, Zeng J, Cao Y, Zeng Y. Recommendation of healthcare services based on an embedded user profile model. Int J Semant Web Inf Syst. 2022;18(1). doi:10.4018/IJSWIS.313198. [Google Scholar] [CrossRef]

24. Hickey BA, Chalmers T, Newton P, Lin CT, Sibbritt D, McLachlan CS, et al. Smart devices and wearable technologies to detect and monitor mental health conditions and stress: a systematic review. Sensors. 2021;21(10):3461. doi:10.3390/s21103461. [Google Scholar] [PubMed] [CrossRef]

25. Hunkin H, King DL, Zajac IT. Perceived acceptability of wearable devices for the treatment of mental health problems. J Clin Psychol. 2020;76(6):987–1003. doi:10.1002/jclp.v76.6. [Google Scholar] [CrossRef]

26. Long N, Lei Y, Peng L, Xu P, Mao P. A scoping review on monitoring mental health using smart wearable devices. Math Biosci Eng. 2022;19(8):7899–919. doi:10.3934/mbe.2022369. [Google Scholar] [PubMed] [CrossRef]

27. Naslund JA, Aschbrenner KA, Bartels SJ. Wearable devices and smartphones for activity tracking among people with serious mental illness. Ment Health Phys Act. 2016;10:10–7. doi:10.1016/j.mhpa.2016.02.001. [Google Scholar] [PubMed] [CrossRef]

28. Olyanasab A, Annabestani M. Leveraging machine learning for personalized wearable biomedical devices: a review. J Pers Med. 2024;14(2):203. doi:10.3390/jpm14020203. [Google Scholar] [PubMed] [CrossRef]

29. Baucas MJ, Spachos P, Plataniotis KN. Federated learning and blockchain-enabled fog-IoT platform for wearables in predictive healthcare. IEEE Trans Comput Soc Syst. 2023;10(4):1732–41. doi:10.1109/TCSS.2023.3235950. [Google Scholar] [CrossRef]

30. Sarivougioukas J, Vagelatos A. Fused contextual data with threading technology to accelerate processing in home UbiHealth. Int J Softw Sci Comput Intell. 2021 Nov. 12;14(1):1–14. doi:10.4018/IJSSCI. [Google Scholar] [CrossRef]

31. Strudwick G, Eyasu T. Electronic health record use by nurses in mental health settings: a literature review. Arch Psychiatr Nurs. 2015;29(4):238–41. doi:10.1016/j.apnu.2015.03.007. [Google Scholar] [PubMed] [CrossRef]

32. Edgcomb JB, Zima B. Machine learning, natural language processing, and the electronic health record: innovations in mental health services research. Psychiatr Serv. 2019;70(4):346–9. doi:10.1176/appi.ps.201800401. [Google Scholar] [PubMed] [CrossRef]

33. Schwarz J, Bärkås A, Blease C, Collins L, Hägglund M, Markham S, et al. Sharing clinical notes and electronic health records with people affected by mental health conditions: scoping review. JMIR Ment Health. 2021;8(12):e34170. doi:10.2196/34170. [Google Scholar] [PubMed] [CrossRef]

34. Larrison CR, Xiang X, Gustafson M, Lardiere MR, Jordan N. Implementation of electronic health records among community mental health agencies. J Behav Health Serv Res. 2018;45:133–42. doi:10.1007/s11414-017-9556-9. [Google Scholar] [PubMed] [CrossRef]

35. Liu Y. Depression clinical detection model based on social media: a federated deep learning approach. J Supercomput. 2024;80(6):7931–54. doi:10.1007/s11227-023-05754-7. [Google Scholar] [CrossRef]

36. Pranto MAM, Al Asad N. A comprehensive model to monitor mental health based on federated learning and deep learning. In: 2021 IEEE International Conference on Signal Processing, Information, Communication & Systems (SPICSCON), 2021; Dhaka, Bangladesh: IEEE; p. 18–21. [Google Scholar]

37. Thummisetti BSP, Atluri H. Advancing healthcare informatics for empowering privacy and security through federated learning paradigms. Int J Sustainable Dev in Comput Sci. 2024;6(1):1–16. [Google Scholar]

38. Onyebuchi A, Matthew UO, Kazaure JS, Okafor NU, Okey OD, Okochi PI, et al. Business demand for a cloud enterprise data warehouse in electronic healthcare computing: issues and developments in e-healthcare cloud computing. Int J Cloud Appl Comput. 2022;12(1):1–22. doi:10.4018/IJCAC.297098. [Google Scholar] [CrossRef]

39. Alahmadi A, Khan HA, Shafiq G, Ahmed J, Ali B, Javed MA, et al. A privacy-preserved IoMT-based mental stress detection framework with federated learning. J Supercomput. 2024;80(8):10255–74. doi:10.1007/s11227-023-05847-3. [Google Scholar] [CrossRef]

40. Gahlan N, Sethia D. Federated learning inspired privacy sensitive emotion recognition based on multi-modal physiological sensors. Cluster Comput. 2024;27(3):3179–201. doi:10.1007/s10586-023-04133-4. [Google Scholar] [CrossRef]

41. Kaushik S, Gandhi C. Capability-based access control with trust for effective healthcare systems. Int J Cloud Appl Comput. 2022;12(1):1–28. [Google Scholar]

42. Tewari A, Gupta BB. An internet-of-things-based security scheme for healthcare environment for robust location privacy. Int J Computat Sci and Eng. 2020;21(2):298. doi:10.1504/IJCSE.2020.105742. [Google Scholar] [CrossRef]

43. Khalil SS, Tawfik NS, Spruit M. Federated learning for privacy-preserving depression detection with multilingual language models in social media posts. Patterns. 2024;5(7):1–9. [Google Scholar]

44. Vasconcelos AB, Drummond LMDA, Brum RC, Paes A. Exploring federated learning to trace depression in social media with language models. In: 2023 International Symposium on Computer Architecture and High Performance Computing Workshops (SBAC-PADW), 2023; Porto Alegre, Brazil; p. 24–30. doi:10.1109/SBAC-PADW60351.2023.00014. [Google Scholar] [CrossRef]

45. Gupta B, Gaurav A, Panigrahi PK. Analysis of security and privacy issues of information management of big data in B2B based healthcare systems. J Bus Res. 2023;162:113859. doi:10.1016/j.jbusres.2023.113859. [Google Scholar] [CrossRef]

46. Quamara M, Gupta BB, Yamaguchi S. An end-to-end security framework for smart healthcare information sharing against botnet-based cyber-attacks. In: 2021 IEEE International Conference on Consumer Electronics (ICCE), 2021; Las Vegas, NV, USA; p. 1–4. doi:10.1109/ICCE50685.2021.9427753. [Google Scholar] [CrossRef]

47. Cui Y, Li Z, Liu L, Zhang J, Liu J. Privacy-preserving speech-based depression diagnosis via federated learning. In: 2022 44th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC), 2022 Jul 11–15; Glasgow, UK; p. 1371–4. [Google Scholar]

48. Bn S, Abdullah S. Privacy sensitive speech analysis using federated learning to assess depression. In: ICASSP 2022–2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2022; Singapore, Singapore; p. 6272–6. doi:10.1109/ICASSP43922.2022.9746827. [Google Scholar] [CrossRef]

49. Qu Z, Chen Z, Dehdashti S, Tiwari P. QFSM: a novel quantum federated learning algorithm for speech emotion recognition with minimal gated unit in 5G IoV. IEEE Trans Intell Vehicles. 2024. [Google Scholar]

50. Zhang T, Feng T, Alam S, Lee S, Zhang M, Narayanan SS, et al. Fedaudio: a federated learning benchmark for audio tasks. In: ICASSP 2023–2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2023; Rhodes Island, Greece; p. 1–5. doi:10.1109/ICASSP49357.2023.10096500. [Google Scholar] [CrossRef]

51. Yin K, Ding Z, Yang X, Tan Z, Zhu R, Ji X, et al. Sleep staging method for imbalanced EEG data based on differential privacy federated learning. Int J on Artif Intell Tools. 2022;31(6):2240018. doi:10.1142/S0218213022400188. [Google Scholar] [CrossRef]

52. Xu C, Liu H, Qi W. EEG emotion recognition based on federated learning framework. Electronics. 2022;11(20):3316. [Google Scholar]

53. Katsigiannis S, Ramzan N. DREAMER: a database for emotion recognition through EEG and ECG signals from wireless low-cost off-the-shelf devices. IEEE J Biomed Health Inform. 2017;22(1):98–107. [Google Scholar] [PubMed]

54. Gore E, Rathi S. Surveying machine learning algorithms on EEG signals data for mental health assessment. In: 2019 IEEE Pune Section International Conference (PuneCon), 2019; Pune, India; p. 1–6. doi:10.1109/PuneCon46936.2019.9105749. [Google Scholar] [CrossRef]

55. Shome D, Kar T. FedAffect: few-shot federated learning for facial expression recognition. In: Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021; Montreal, QC, Canada; p. 4168–75. [Google Scholar]

56. Chhikara P, Singh P, Tekchandani R, Kumar N, Guizani M. Federated learning meets human emotions: a decentralized framework for human-computer interaction for IoT applications. IEEE Internet Things J. 2020;8(8):6949–62. [Google Scholar]

57. Salman A, Busso C. Privacy preserving personalization for video facial expression recognition using federated learning. In: Proceedings of the 2022 International Conference on Multimodal Interaction, 2022; Bengaluru, India; p. 495–503. [Google Scholar]

58. Che L, Wang J, Zhou Y, Ma F. Multimodal federated learning: a survey. Sensors. 2023;23(15):6986. doi:10.3390/s23156986. [Google Scholar] [PubMed] [CrossRef]

59. Gahlan N, Sethia D. AFLEMP: attention-based federated learning for emotion recognition using multi-modal physiological data. Biomed Signal Process Control. 2024;94:106353. doi:10.1016/j.bspc.2024.106353. [Google Scholar] [CrossRef]

60. Geetha AV, Mala T, Priyanka D, Uma E. Multimodal emotion recognition with deep learning: advancements, challenges, and future directions. Inf Fusion. 2024;105:102218. doi:10.1016/j.inffus.2023.102218. [Google Scholar] [CrossRef]

61. Gratch J, Artstein R, Lucas GM, Stratou G, Scherer S, Nazarian A, et al. The distress analysis interview corpus of human and computer interviews. In: Proceedings of the Ninth International Conference on Language Resources and Evaluation, 2014; p. 3123–8. [Google Scholar]

62. Schmidt P, Reiss A, Duerichen R, Marberger C, Van Laerhoven K. Introducing wesad, a multimodal dataset for wearable stress and affect detection. In: Proceedings of the 20th ACM International Conference on Multimodal Interaction, 2018; Boulder, CO, USA; p. 400–8. [Google Scholar]

63. Fauzi MA, Yang B, Blobel B. Comparative analysis between individual, centralized, and federated learning for smartwatch based stress detection. J Pers Med. 2022;12(10):1584. doi:10.3390/jpm12101584. [Google Scholar] [PubMed] [CrossRef]

64. Koelstra S, Muhl C, Soleymani M, Lee JS, Yazdani A, Ebrahimi T, et al. DEAP: a database for emotion analysis; using physiological signals. IEEE Trans Affect Comput. 2011;3(1):18–31. doi:10.1109/T-AFFC.2011.15. [Google Scholar] [CrossRef]

65. Nandi A, Xhafa F. A federated learning method for real-time emotion state classification from multi-modal streaming. Methods. 2022;204:340–7. doi:10.1016/j.ymeth.2022.03.005. [Google Scholar] [PubMed] [CrossRef]

66. Agrawal M, Anwar MA, Jindal R. Fedcer-emotion recognition using 2D-CNN in decentralized federated learning environment. In: 2023 6th International Conference on Information Systems and Computer Networks (ISCON), 2023; Mathura, India; p. 1–5. [Google Scholar]

67. Miranda-Correa JA, Abadi MK, Sebe N, Patras I. AMIGOS: a dataset for affect, personality and mood research on individuals and groups. IEEE Trans Affect Comput. 2018;12(2):479–93. doi:10.1109/TAFFC.2018.2884461. [Google Scholar] [CrossRef]

68. Busso C, Bulut M, Lee CC, Kazemzadeh A, Mower E, Kim S, et al. IEMOCAP: interactive emotional dyadic motion capture database. Lang Resour Eval. 2008;42:335–59. doi:10.1007/s10579-008-9076-6. [Google Scholar] [CrossRef]

69. Latif S, Khalifa S, Rana R, Jurdak R. Federated learning for speech emotion recognition applications. In: 2020 19th ACM/IEEE International Conference on Information Processing in Sensor Networks (IPSN), 2020; Sydney, NSW, Australia; p. 341–2. [Google Scholar]

70. El Haouij N, Poggi JM, Sevestre-Ghalila S, Ghozi R, Jaidane M. AffectiveROAD system and database to assess driver’s attention. In: Proceedings of the 33rd Annual ACM Symposium on Applied Computing, 2018; Pau, France; p. 800–3. [Google Scholar]

71. Goodfellow IJ, Erhan D, Carrier PL, Courville A, Mirza M, Hamner B, et al. Challenges in representation learning: a report on three machine learning contests. In: Neural Information Processing: 20th International Conference, ICONIP 2013, 2013 Nov 3–7; Daegu, Republic of Korea; p. 117–24. [Google Scholar]

72. Srihitha PVNP, Verma M, Prasad MVNK. Federated scaling of pre-trained models for deep facial expression recognition. In: International Conference on Computer Vision and Image Processing, 2023; Brussels, Belgium; p. 90–101. [Google Scholar]

73. Jaswanth M, Narayana NKL, Rahul S, Aiswariya Milan R, Amudha J. Emotion and advertising effectiveness: a novel facial expression analysis approach using federated learning. In: 2023 IEEE 20th India Council International Conference (INDICON), 2023; Hyderabad, India; p. 368–73. [Google Scholar]

74. Livingstone SR, Russo FA. The ryerson audio-visual database of emotional speech and song (RAVDESSa dynamic, multimodal set of facial and vocal expressions in North American English. PLoS One. 2018;13(5):e0196391. doi:10.1371/journal.pone.0196391. [Google Scholar] [PubMed] [CrossRef]

75. Chawla M, Panda SN, Khullar V, Kumar S, Bhattacharjee SB. A lightweight and privacy preserved federated learning ecosystem for analyzing verbal communication emotions in identical and non-identical databases. Meas Sens. 2024;34:101268. doi:10.1016/j.measen.2024.101268. [Google Scholar] [CrossRef]

76. Murthy BSNVR, Kishore VR. Federated learning approach based on the MFCC for speech emotion recognition. In: Algorithms in Advanced Artificial Intelligence: ICAAAI-2023, 2024; Andhra Pradesh, India; p. 77. [Google Scholar]

77. Raju VVN, Saravanakumar R, Yusuf N, Pradhan R, Hamdi H, Saravanan KA, et al. Enhancing emotion prediction using deep learning and distributed federated systems with SMOTE oversampling technique. Alex Eng J. 2024;108:498–508. doi:10.1016/j.aej.2024.07.081. [Google Scholar] [CrossRef]

78. Zheng WL, Liu W, Lu Y, Lu BL, Cichocki A. Emotionmeter: a multimodal framework for recognizing human emotions. IEEE Trans Cybern. 2018;49(3):1110–22. [Google Scholar] [PubMed]

79. Liu Y, Yang R. Federated learning application on depression treatment robots (DTbot). In: 2021 IEEE 13th International Conference on Computer Research and Development (ICCRD), 2021; Beijing, China; p. 121–4. [Google Scholar]

80. Xu X, Peng H, Bhuiyan MZA, Hao Z, Liu L, Sun L, et al. Privacy-preserving federated depression detection from multisource mobile health data. IEEE Trans Industr Inform. 2021;18(7):4788–97. [Google Scholar]

81. Lemos R, Areias-Marques S, Ferreira P, O’Brien P, Beltrán-Jaunsarás ME, Ribeiro G, et al. A prospective observational study for a Federated Artificial Intelligence solution for moniToring mental Health status after cancer treatment (FAITHstudy protocol. BMC Psychiatry. 2022;22(1):817. doi:10.1186/s12888-022-04446-5. [Google Scholar] [PubMed] [CrossRef]

82. Gupta A, Maurya C, Dhere K, Chaurasiya VK. Wellness detection using clustered federated learning. In: 2022 IEEE 6th Conference on Information and Communication Technology (CICT), 2022; Gwalior, India; p. 1–5. [Google Scholar]

83. Lee DY, Choi B, Kim C, Fridgeirsson E, Reps J, Kim M, et al. Privacy-preserving federated model predicting bipolar transition in patients with depression: prediction model development study. J Med Internet Res. 2023;25:e46165. doi:10.2196/46165. [Google Scholar] [PubMed] [CrossRef]

84. Gupta C, Khullar V. Privacy preserving collaboratively training framework for classification of major depressive disorder using non-IID three channel electroencephalogram. Procedia Comput Sci. 2024;235:56–69. doi:10.1016/j.procs.2024.04.006. [Google Scholar] [CrossRef]

85. Vyas J, Das D, Das SK. Vehicular edge computing based driver recommendation system using federated learning. In: 2020 IEEE 17th International Conference on Mobile Ad Hoc and Sensor Systems (MASS), 2020; Delhi, India; p. 675–83. [Google Scholar]

86. Rafi H, Benezeth Y, Reynaud P, Arnoux E, Song FY, Demonceaux C. Personalization of AI models based on federated learning for driver stress monitoring. In: European Conference on Computer Vision, 2022; Tel Aviv, Israel; p. 575–85. [Google Scholar]

87. Jiang S, Firouzi F, Chakrabarty K. Low-overhead clustered federated learning for personalized stress monitoring. IEEE Internet Things J. 2024;11(3):4335–47. doi:10.1109/JIOT.2023.3299736. [Google Scholar] [CrossRef]

88. Almadhor A, Sampedro GA, Abisado M, Abbas S, Kim YJ, Khan MA, et al. Wrist-based electrodermal activity monitoring for stress detection using federated learning. Sensors. 2023;23(8):3984. doi:10.3390/s23083984. [Google Scholar] [PubMed] [CrossRef]

89. Can YS, Ersoy C. Privacy-preserving federated deep learning for wearable IoT-based biomedical monitoring. ACM Trans Internet Technol. 2021;21(1):1–17. doi:10.1145/3428152. [Google Scholar] [CrossRef]

90. Vyas J, Bhumika, Das D, Chaudhury S. Federated learning based driver recommendation for next generation transportation system. Expert Syst Appl. 2023;225:119951. doi:10.1016/j.eswa.2023.119951. [Google Scholar] [CrossRef]

91. Khoa TA, Nguyen DV, Van Nguyen Thi P, Zettsu K. FedMCRNN: federated learning using multiple convolutional recurrent neural networks for sleep quality prediction. In: Proceedings of the 3rd Acm Workshop on Intelligent Cross-Data Analysis and Retrieval, 2022; Newark, NJ, USA; p. 63–9. [Google Scholar]

92. Anido-Alonso A, Alvarez-Estevez D. Decentralized data-privacy preserving deep-learning approaches for enhancing inter-database generalization in automatic sleep staging. IEEE J Biomed Health Inform. 2023; 27(11):5610–21. [Google Scholar] [PubMed]

93. Ma B, Duan L, Huang Z, Qiao Y, Gong B. A federated semi-supervised automatic sleep staging method based on relationship knowledge sharing. Expert Syst Appl. 2024;237:121427. doi:10.1016/j.eswa.2023.121427. [Google Scholar] [CrossRef]

94. Borges GA, dos Anjos JCS, Silva JS. Federated learning for sleep detection problems. In: 2024 IEEE International Conference on Smart Computing (SMARTCOMP), 2024; Osaka, Japan; p. 368–73. [Google Scholar]

95. Dautov R, Husom EJ. Drowsiness detection using federated learning: lessons learnt from dealing with non-IID data. In: Proceedings of the 17th International Conference on PErvasive Technologies Related to Assistive Environments, 2024; Crete, Greece; p. 285–92. [Google Scholar]

96. Simić N, Suzić S, Milošević N, Stanojev V, Nosek T, Popović B, et al. Enhancing emotion recognition through federated learning: a multimodal approach with convolutional neural networks. Appl Sci. 2024;14(4):1325. doi:10.3390/app14041325. [Google Scholar] [CrossRef]

97. Singh AK, Kumar A, Choi BJ. Privacy-preserving digital intervention for mental health using federated learning. In: International Conference on Intelligent Human Computer Interaction, 2022; Tashkent, Uzbekistan; p. 213–24. [Google Scholar]

98. Zhang C, Li M, Wu D. Federated multidomain learning with graph ensemble autoencoder GMM for emotion recognition. IEEE Trans Intell Transp Syst. 2022;24(7):7631–41. [Google Scholar]

99. Chen C. Research on online teaching emotion detection based on federated learning. In: Proceedings of the 2023 3rd International Conference on Bioinformatics and Intelligent Computing, 2023; Sanya, China; p. 159–64. [Google Scholar]

100. Qirtas MM, Pesch D, Zafeiridi E, White EB. Privacy preserving loneliness detection: a federated learning approach. In: 2022 IEEE International Conference on Digital Health (ICDH), 2022; Barcelona, Spain; p. 157–62. [Google Scholar]

101. Ahmed U, Lin JCW, Srivastava G. Hyper-graph attention based federated learning methods for use in mental health detection. IEEE J Biomed Health Inform. 2022;27(2):768–77. [Google Scholar]

102. Sahu SK, Azam M, Takkar M, Bansal J, Patra JP, Kunal. Federated learning for private AI diagnosis of schizophrenia. In: Federated learning and privacy-preserving in healthcare AI. USA: IGI Global; 2024. p. 137–57. [Google Scholar]

103. Aversa R, Bochicchio M, Branco D, Magliulo M, Orlando A, Pristner A, et al. REDRAW: fedeRatED learning for human wellbeing. In: International Conference on Advanced Information Networking and Applications, 2024; Kitakyushu, Japan; p. 101–9. [Google Scholar]

104. Shen S, Zhu T, Wu D, Wang W, Zhou W. From distributed machine learning to federated learning: in the view of data privacy and security. Concurr Comput. 2022;34(16):e6002. doi:10.1002/cpe.v34.16. [Google Scholar] [CrossRef]

105. Moshawrab M, Adda M, Bouzouane A, Ibrahim H, Raad A. Reviewing federated machine learning and its use in diseases prediction. Sensors. 2023;23(4):2112. doi:10.3390/s23042112. [Google Scholar] [PubMed] [CrossRef]

106. Ogundokun RO, Misra S, Maskeliunas R, Damasevicius R. A review on federated learning and machine learning approaches: categorization, application areas, and blockchain technology. Information. 2022;13(5):263. doi:10.3390/info13050263. [Google Scholar] [CrossRef]

107. AbdulRahman S, Tout H, Ould-Slimane H, Mourad A, Talhi C, Guizani M. A survey on federated learning: the journey from centralized to distributed on-site learning and beyond. IEEE Internet Things J. 2020;8(7):5476–97. [Google Scholar]

108. Bellavista P, Foschini L, Mora A. Decentralised learning in federated deployment environments: a system-level survey. ACM Comput Surveys. 2021;54(1):1–38. [Google Scholar]

109. Imteaj A, Thakker U, Wang S, Li J, Amini MH. A survey on federated learning for resource-constrained IoT devices. IEEE Internet Things J. 2021;9(1):1–24. [Google Scholar]

110. Liu G, Wang C, Ma X, Yang Y. Keep your data locally: federated-learning-based data privacy preservation in edge computing. IEEE Netw. 2021;35(2):60–6. doi:10.1109/MNET.011.2000215. [Google Scholar] [CrossRef]

111. Lim WYB, Luong NC, Hoang DT, Jiao Y, Liang YC, Yang Q, et al. Federated learning in mobile edge networks: a comprehensive survey. IEEE Commun Surv Tut. 2020;22(3):2031–63. doi:10.1109/COMST.2020.2986024. [Google Scholar] [CrossRef]

112. Mothukuri V, Parizi RM, Pouriyeh S, Huang Y, Dehghantanha A, Srivastava G. A survey on security and privacy of federated learning. Future Gener Comput Syst. 2021;115:619–40. doi:10.1016/j.future.2020.10.007. [Google Scholar] [CrossRef]

113. Bouacida N, Mohapatra P. Vulnerabilities in federated learning. IEEE Access. 2021;9:63229–49. doi:10.1109/ACCESS.2021.3075203. [Google Scholar] [CrossRef]

114. Qi P, Chiaro D, Guzzo A, Ianni M, Fortino G, Piccialli F. Model aggregation techniques in federated learning: a comprehensive survey. Future Gener Comput Syst. 2023;150:272–93. [Google Scholar]

115. Zhong J, Wu Y, Ma W, Deng S, Zhou H. Optimizing multi-objective federated learning on non-IID data with improved NSGA-III and hierarchical clustering. Symmetry. 2022;14(5):1070. doi:10.3390/sym14051070. [Google Scholar] [CrossRef]

116. Duan M, Liu D, Chen X, Tan Y, Ren J, Qiao L, et al. Astraea: self-balancing federated learning for improving classification accuracy of mobile deep learning applications. In: 2019 IEEE 37th International Conference on Computer Design (ICCD), 2019; Abu Dhabi, United Arab Emirates; p. 246–54. [Google Scholar]

117. Gafni T, Shlezinger N, Cohen K, Eldar YC, Poor HV. Federated learning: a signal processing perspective. IEEE Signal Process Mag. 2022;39(3):14–41. doi:10.1109/MSP.2021.3125282. [Google Scholar] [CrossRef]

118. Chen M, Shlezinger N, Poor HV, Eldar YC, Cui S. Communication-efficient federated learning. Proc Nat Acad Sci. 2021;118(17):e2024789118. doi:10.1073/pnas.2024789118. [Google Scholar] [PubMed] [CrossRef]

119. Wahab OA, Mourad A, Otrok H, Taleb T. Federated machine learning: survey, multi-level classification, desirable criteria and future directions in communication and networking systems. IEEE Commun Surv Tut. 2021;23(2):1342–97. doi:10.1109/COMST.2021.3058573. [Google Scholar] [CrossRef]

120. Gao Y, Parcollet T, Zaiem S, Fernandez-Marques J, de Gusmao PPB, Beutel DJ, et al. End-to-end speech recognition from federated acoustic models. In: ICASSP 2022–2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2022; Singapore, Singapore; p. 7227–31. [Google Scholar]

121. Guliani D, Beaufays F, Motta G. Training speech recognition models with federated learning: a quality/cost framework. In: ICASSP 2021–2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2021; Toronto, ON, Canada; p. 3080–4. [Google Scholar]

122. Nagy B, Hegedűs I, Sándor N, Egedi B, Mehmood H, Saravanan K, et al. Privacy-preserving federated learning and its application to natural language processing. Knowl-Based Syst. 2023;268:110475. doi:10.1016/j.knosys.2023.110475. [Google Scholar] [CrossRef]

123. KhoKhar FA, Shah JH, Khan MA, Sharif M, Tariq U, Kadry S. A review on federated learning towards image processing. Comput and Electr Eng. 2022;99:107818. doi:10.1016/j.compeleceng.2022.107818. [Google Scholar] [CrossRef]

124. Samee NA, Khan U, Khan S, Jamjoom MM, Sharif M, Kim DH. Safeguarding online spaces: a powerful fusion of federated learning, word embeddings, and emotional features for cyberbullying detection. IEEE Access. 2023;11:124524–41. [Google Scholar]

125. Padate R, Kalla M, Gupta A, Sharma A. Federated learning for image captioning: a comprehensive review of privacy-preserving collaborative model training in distributed environments. In: 2023 2nd International Conference on Edge Computing and Applications (ICECAA), 2023; Namakkal, India; p. 857–65. [Google Scholar]

126. Ran AR, Wang X, Chan PP, Wong MOM, Yuen H, Lam NM, et al. Developing a privacy-preserving deep learning model for glaucoma detection: a multicentre study with federated learning. Br J Ophthalmol. 2024;108(8):1114–23. doi:10.1136/bjo-2023-324188. [Google Scholar] [PubMed] [CrossRef]

127. Beutel DJ, Topal T, Mathur A, Qiu X, Fernandez-Marques J, Gao Y, et al. Flower: a friendly federated learning research framework. arXiv preprint arXiv:200714390. 2020. [Google Scholar]

128. Ziller A, Trask A, Lopardo A, Szymkow B, Wagner B, Bluemke E, et al. PySyft: a library for easy federated learning. In: Federated Learning Systems: Towards Next-Generation AI. Cham, New York: Springer; 2021. p. 111–39. [Google Scholar]

129. Siddiqui MSB, Shusmita SA, Sabreen S, Alam MGR. FedNet: federated implementation of neural networks for facial expression recognition. In: 2022 International Conference on Decision Aid Sciences and Applications (DASA), 2022; Chiangrai, Thailand; p. 82–7. [Google Scholar]

130. Campos EM, Saura PF, González-Vidal A, Hernández-Ramos JL, Bernabe JB, Baldini G, et al. Evaluating federated learning for intrusion detection in internet of things: review and challenges. Comput Netw. 2022;203:108661. doi:10.1016/j.comnet.2021.108661. [Google Scholar] [CrossRef]

131. Lee GH, Shin SY. Federated learning on clinical benchmark data: performance assessment. J Med Internet Res. 2020;22(10):e20891. doi:10.2196/20891. [Google Scholar] [PubMed] [CrossRef]

132. Lee H, Bertozzi AL, Kovacevic J, Chi Y. Privacy-preserving federated multi-task linear regression: a one-shot linear mixing approach inspired by graph regularization. In: ICASSP 2022–2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2022; Singapore, Singapore: IEEE; p. 5947–51. [Google Scholar]

133. Almanifi ORA, Chow CO, Tham ML, Chuah JH, Kanesan J. Communication and computation efficiency in federated learning: a survey. Internet Things. 2023 Jul;22:100742. [Google Scholar]

134. Zaabar B, Cheikhrouhou O, Jamil F, Ammi M, Abid M. HealthBlock: a secure blockchain-based healthcare data management system. Comput Netw. 2021;200:108500. doi:10.1016/j.comnet.2021.108500. [Google Scholar] [CrossRef]

135. Li Q, Wen Z, Wu Z, Hu S, Wang N, Li Y, et al. A survey on federated learning systems: vision, hype and reality for data privacy and protection. IEEE Trans Knowl Data Eng. 2021;35(4):3347–66. [Google Scholar]

136. Truong N, Sun K, Wang S, Guitton F, Guo Y. Privacy preservation in federated learning: an insightful survey from the GDPR perspective. Comput Secur. 2021;110:102402. doi:10.1016/j.cose.2021.102402. [Google Scholar] [CrossRef]

137. Hakak S, Ray S, Khan WZ, Scheme E. A framework for edge-assisted healthcare data analytics using federated learning. In: 2020 IEEE International Conference on Big Data (Big Data), 2020; Atlanta, GA, USA; p. 3423–7. [Google Scholar]

138. Wagan SA, Koo J, Siddiqui IF, Attique M, Shin DR, Qureshi NMF. Internet of medical things and trending converged technologies: a comprehensive review on real-time applications. J King Saud Univ Comput Inf Sci. 2022;34(10):9228–51. [Google Scholar]

139. Arikumar KS, Prathiba SB, Alazab M, Gadekallu TR, Pandya S, Khan JM, et al. FL-PMI: federated learning-based person movement identification through wearable devices in smart healthcare systems. Sensors. 2022;22(4):1377. doi:10.3390/s22041377. [Google Scholar] [PubMed] [CrossRef]

140. Handa T, Singhal I, Chakraborty P, Kaur G. Recent trends of federated learning for smart healthcare systems. In: Federated learning and AI for Healthcare 5.0. USA: IGI Global; 2024. p. 78–103. [Google Scholar]

141. Yang Q. Toward responsible AI: an overview of federated learning for user-centered privacy-preserving computing. ACM Trans on Interact Intell Syst. 2021;11(3–4):1–22. [Google Scholar]

142. Mendieta M, Yang T, Wang P, Lee M, Ding Z, Chen C. Local learning matters: rethinking data heterogeneity in federated learning. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022; New Orleans Ernest N. Morial Convention Center, New Orleans, LA, USA; p. 8397–406. [Google Scholar]

143. Yang M, Wang X, Zhu H, Wang H, Qian H. Federated learning with class imbalance reduction. In: 2021 29th European Signal Processing Conference (EUSIPCO), 2021; Dublin, Ireland; p. 2174–8. [Google Scholar]

144. Qin Z, Yang L, Wang Q, Han Y, Hu Q. Reliable and interpretable personalized federated learning. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023; Vancouver, BC, Canada; p. 20422–31. [Google Scholar]

145. Daole M, Ducange P, Marcelloni F, Renda A. Trustworthy AI in heterogeneous settings: federated learning of explainable classifiers. In: 2024 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE), 2024; Yokohama, Japan; p. 1–9. [Google Scholar]

146. Wang Z, Yang Z, Azimi I, Rahmani AM. Differential private federated transfer learning for mental health monitoring in everyday settings: a case study on stress detection. arXiv preprint arXiv:2402.10862. 2024. [Google Scholar]

147. Chen Z, Tian P, Liao W, Yu W. Towards multi-party targeted model poisoning attacks against federated learning systems. High-Confidence Comput. 2021;1(1):100002. doi:10.1016/j.hcc.2021.100002. [Google Scholar] [CrossRef]

148. Tariq A, Serhani MA, Sallabi FM, Barka ES, Qayyum T, Khater HM, et al. Trustworthy federated learning: a comprehensive review, architecture, key challenges, and future research prospects. IEEE Open J Commun Soc. 2024;5:4920–98. [Google Scholar]

149. Zhang Y, Zeng D, Luo J, Fu X, Chen G, Xu Z, et al. A survey of trustworthy federated learning: issues, solutions, and challenges. ACM Trans Intell Syst Technol. 2024;15(6):47. [Google Scholar]

150. Arundel R, Li A, Baker E, Bentley R. Housing unaffordability and mental health: dynamics across age and tenure. Int J Hous Policy. 2024;24(1):44–74. doi:10.1080/19491247.2022.2106541. [Google Scholar] [CrossRef]

151. Karnbach K, Witkowski M, Ebrahimi OV, Burger J. Back to life, back to reality: a multi-level dynamic network analysis of student mental health upon return to campus during the COVID-19 pandemic. Curr Psychol. 2024;43(14):12998–3010. doi:10.1007/s12144-022-03196-7. [Google Scholar] [PubMed] [CrossRef]

152. Nelson B, McGorry PD, Wichers M, Wigman JTW, Hartmann JA. Moving from static to dynamic models of the onset of mental disorder: a review. JAMA Psychiatr. 2017;74(5):528–34. doi:10.1001/jamapsychiatry.2017.0001. [Google Scholar] [PubMed] [CrossRef]

153. Peyvandi A, Majidi B, Peyvandi S, Patra JC. Privacy-preserving federated learning for scalable and high data quality computational-intelligence-as-a-service in Society 5.0. Multimed Tools Appl. 2022;81(18):25029–50. doi:10.1007/s11042-022-12900-5. [Google Scholar] [PubMed] [CrossRef]

154. Thapa C, Camtepe S. Precision health data: requirements, challenges and existing techniques for data security and privacy. Comput Biol Med. 2021;129:104130. doi:10.1016/j.compbiomed.2020.104130. [Google Scholar] [PubMed] [CrossRef]

155. Armontrout J, Torous J, Fisher M, Drogin E, Gutheil T. Mobile mental health: navigating new rules and regulations for digital tools. Curr Psychiat Rep. 2016;18:1–7. [Google Scholar]

156. Holzinger A, Dehmer M, Emmert-Streib F, Cucchiara R, Augenstein I, Del Ser J, et al. Information fusion as an integrative cross-cutting enabler to achieve robust, explainable, and trustworthy medical artificial intelligence. Inf Fusion. 2022;79:263–78. doi:10.1016/j.inffus.2021.10.007. [Google Scholar] [CrossRef]

157. Dou Q, So TY, Jiang M, Liu Q, Vardhanabhuti V, Kaissis G, et al. Federated deep learning for detecting COVID-19 lung abnormalities in CT: a privacy-preserving multinational validation study. npj Digit Med. 2021;4(1):60. doi:10.1038/s41746-021-00431-6. [Google Scholar] [PubMed] [CrossRef]