Open Access

ARTICLE

A Deep Learning Based Sentiment Analytic Model for the Prediction of Traffic Accidents

Nadeem Malik^1,*, Saud Altaf¹, Muhammad Usman Tariq², Ashir Ahmed³, Muhammad Babar⁴

1 University Institute of Information Technology, PMAS Arid Agriculture University, Rawalpindi, 46000, Pakistan
2 Department of Marketing, Operations, and Information System, Abu Dhabi University, Abu Dhabi, 6844, United Arab Emirates
3 College of Computer and Information Sciences, Prince Sultan University, Riyadh, 11442, Saudi Arabia
4 Robotics and Internet of Things Lab, Prince Sultan University, Riyadh, 11442, Saudi Arabia

* Corresponding Author: Nadeem Malik. Email:

(This article belongs to the Special Issue: The Next Generation of Artificial Intelligence and the Intelligent Internet of Things)

Computers, Materials & Continua 2023, 77(2), 1599-1615. https://doi.org/10.32604/cmc.2023.040455

Received 19 March 2023; Accepted 02 August 2023; Issue published 29 November 2023

Abstract

The severity of traffic accidents is a serious global concern, particularly in developing nations. Knowing the main causes and contributing circumstances may reduce the severity of traffic accidents. There exist many machine learning models and decision support systems to predict road accidents by using datasets from different social media forums such as Twitter, blogs and Facebook. Although such approaches are popular, there exists an issue of data management and low prediction accuracy. This article presented a deep learning-based sentiment analytic model known as Extra-large Network Bi-directional long short term memory (XLNet-Bi-LSTM) to predict traffic collisions based on data collected from social media. Initially, a Tweet dataset has been formed by using an exhaustive keyword-based searching strategy. In the next phase, two different types of features named as individual tokens and pair tokens have been obtained by using POS tagging and association rule mining. The output of this phase has been forwarded to a three-layer deep learning model for final prediction. Numerous experiment has been performed to test the efficiency of the proposed XLNet-Bi-LSTM model. It has been shown that the proposed model achieved 94.2% prediction accuracy.

---

Keywords

---