DOI:10.32604/cmc.2021.013327
| Computers, Materials & Continua DOI:10.32604/cmc.2021.013327 | |
| Article |
Geospatial Analytics for COVID-19 Active Case Detection
1Multimedia University, Cyberjaya, 63100, Malaysia
2AIME Healthcare Sdn Bhd, Kuala Lumpur, 59200, Malaysia
*Corresponding Author: Choo-Yee Ting. Email: cyting@mmu.edu.my
Received: 02 August 2020; Accepted: 12 November 2020
Abstract: Ever since the COVID-19 pandemic started in Wuhan, China, much research work has been focusing on the clinical aspect of SARS-CoV-2. Researchers have been leveraging on various Artificial Intelligence techniques as an alternative to medical approach in understanding the virus. Limited studies have, however, reported on COVID-19 transmission pattern analysis, and using geography features for prediction of potential outbreak sites. Predicting the next most probable outbreak site is crucial, particularly for optimizing the planning of medical personnel and supply resources. To tackle the challenge, this work proposed distance-based similarity measures to predict the next most probable outbreak site together with its magnitude, when would the outbreak likely to happen and the duration of the outbreak. The work began with preprocessing of 1365 patient records from six districts in the most populated state named Selangor in Malaysia. The dataset was then aggregated with population density information and human elicited geography features that might promote the transmission of COVID-19. Empirical findings indicated that the proposed unified decision-making approach outperformed individual distance metric in predicting the total cases, next outbreak location, and the time interval between start dates of two similar sites. Such findings provided valuable insights for policymakers to perform Active Case Detection.
Keywords: COVID-19; geospatial analytics; active case detection
The world is currently experiencing COVID-19 pandemic, which had started at Wuhan, China in December 2019. Despite the reduction in daily confirmed cases in China, outside of China has shown a drastic increase in daily confirmed cases. Countries like the United States, Brazil, and India recorded a daily increase of more than 10000 confirmed cases [1].
In South-East Asia, Malaysia had experienced its first three COVID-19 confirmed cases on 25 January 2020 [2]. Subsequent cases were reported, and many of them were imported cases, including those from China and the United States [3]. In late February, the number of COVID-19 confirmed cases increased when there were Malaysians who travelled back from China during that time. In March 2020, the Tabligh event that had involved almost 12000 participants was then resorted to the high daily growth rate in Malaysia, particularly in the state of Selangor [4]. Due to the exponential growth in confirmed cases, Movement Control Order (MCO) was introduced on 18 March 2020, which was later extended to its 2nd phase on 1 April 2020 and 3rd phase on 15 April 2020 and subsequently entered into its 4th phase of MCO which ends on 12 May 2020 [5]. Currently, Malaysia is in its Recovery MCO (RMCO) until 31 August 2020. During the two MCOs, both the federal and state governments have implemented various interventions to prevent a further outbreak by identifying close contacts and conducting mass testing for Active Case Detection (ACD). MCO is a measure to keep the healthcare system from being overwhelmed by “flattening the curve” [6] and to minimize the negative impact to all economic sectors caused by the prolong MCO. One of the many approaches initiated by the Selangor state government was to deploy geospatial analytics to identify the next most probable areas with an outbreak and estimate the magnitudes.
Geospatial analytics has been employed to tackle challenges in site selection for various business domains. This has included the retail business [7–12], real-estate [13], public safety [14], disaster monitoring and prevention [12], government [14], planetary [15], agriculture [16], and renewal energy [17,18]. Researchers have employed different machine learning methods in geospatial analytics. One of the most common methods that incorporate human knowledge is through Analytical Hierarchical Technique [18–22]. Other researchers employed Decision Support Systems [23] and Fuzzy Logic [24].
Researchers have recently attempted Geographic Information Systems (GIS) for tackling challenges related to COVID-19 and utilization of Big Data to provide insights in decision-making [25–27]. Another recent study by Mollalo and colleagues highlighted the use of GIS in modelling the incident and spread of COVID19 in the United States [28]. The utilization of GIS has also been reported in Iran, where epidemiological maps of cases were developed to monitor the incident locations and rates. In Malaysia, geospatial analytics has been deployed for Active Case Detection (ACD) in late April 2020 in Selangor, Malaysia. However, successful ACD requires tackling the following challenges:
i) how to identify the next most probable outbreak sites (i.e., residential areas)?
ii) how to estimate the magnitude of COVID-19 cases for a particular site (i.e., residential area)?
iii) what is the time interval between the current and the subsequent case, given two similar sites (i.e., residential area)?
In this paper, the discussion has focused on COVID-19 outbreak that happened at a residential area, and therefore the “site” refers residential area. Tackling the above challenges began with gathering, preprocessing and transformation of raw datasets (see Section 3) into an analytical dataset (see Section 4.1) before different distance metrics can be applied. A detailed discussion about the raw datasets and their transformation can be found in Section 4. In Section 5, findings of different distance metrics together with their comparison with the proposed unified decision-making mechanism will be reported.
A similarity or dissimilarity measure is a real-valued function that quantifies the similarity between two items, which in this context refer to records in a database. There are various similarity measure techniques proposed by researchers [29], and they can be categorized into (i) distance-based similarity measure, (ii) feature-based similarity measure, and (iii) probabilistic similarity measure. Among the three techniques, the distance-based similarity measure metric remains most commonly applied in both the academic and commercial settings. A proper distance requires a function d to follow the following properties:
i) positivity: 
ii) symmetry: d(x, y) = d(y, x)
iii) identity-discerning: 
iv) triangle inequality: 
The commonly applied point-based distance measures are the Euclidean distance, City Block distance, and Bray Curtis [30]. Both Euclidean distance and City Block are a special case of the Minkowski family [31–34]. The Minkowski distance can be defined by
where m is a positive real number while xi and yi are two vectors in n-dimension space. By instantiating m = 1, Eq. (1) is modified to take new form as shown below, and it is known as the City Block distance (also referred to as the Manhattan distance). The distance between two items is the sum of the differences of their corresponding components. The City Block distance fulfils the positivity property in which it is always greater than or equal to zero, as well as identity-discerning property where the measurement would be high for points that show dissimilarity and zero for identical points.
When Eq. (1) is instantiated by m = 2, the Euclidean distance can then be defined by
Euclidean inherited the characteristics of Minkowski, and therefore with the existence of large-scale continuous values, the large-scale attributes would dominate the others [35–37]. To overcome such challenges, researchers introduced normalization of continuous features as a solution to the issue [36]. By introducing the coefficient 1/2 the output of Euclidean distance is scaled down (normalized) so that a comparison between two points can be accurately presented. Researchers had proposed an improved version of Euclidean distance such as the Weighted Euclidean [38]. It is employed to overcome the large-scale attribute issue by the Euclidean distance (Eq. 2), which incorporates feature weights w1, w2, 
on each dimension. While having different weights is crucial, redundant weight assignment could dominate the similarity between data points [39]. Due to incomplete information about influence of geography characteristics towards COVID-19, this study firstly normalized the continuous data before applying Euclidean distance metric.
Bray Curtis distance measure [30] quantifies the difference between two samples. The dissimilarity is computed on the raw sum of each sample and the sum of differences between the features of each sample. Similarity can be obtained if the distance or dissimilarity is subtracted by 100.
In this study, only point-based distance metrics were considered. This is because the number of variables that formed the analytical dataset is fixed, and the distance (magnitude) between two points is crucial for prediction work to addresses the challenges successfully.
This section begins with a discussion about the raw datasets and the relevant transformation process to form an analytical dataset. Subsequently, the discussion will be centred around the implementation of different distance metrics onto the analytical dataset to answer the research questions.
Tab. 1 shows the different datasets used in this research work. The first dataset, De, consists of 1365 patient addresses, with many of them incomplete addresses [40]. The second dataset, Dpop, consists of raw population density about children under five, elderly over 60, men, woman, women of reproductive age, and youth [41]. Preprocessing on the raw dataset was needed because the dataset provides only population density estimation at the level of individual latitude and longitude. In this study, the population density was calculated using 500 m radius for each patient location.
The third dataset, Dres, consists of 3374 residential areas in the Selangor state. It originated from the Valuation and Property Services Department Malaysia [42] and has been digitized by Brickz.my [43]. The purpose is to detect the names and types of residential area the patients live.
The fourth dataset, Dpoi, comprises most of the points of interest(POI) in the Selangor state. The dataset was subscribed from Telekom Malaysia [44], and it has 116402 POI grouped into 1116 categories. Examples of POI are Burger King, KFC, McDonald, Ace Hardware, Starbucks, Ikea, Parkson, and many more. In the dataset, each point of interest is tagged to a particular category. Examples of the category are Food & Beverages, Convenient Store, Wet Market, Hardware Store, Hypermarket, and many more.
Dcase contains daily COVID-19 confirmed cases in Selangor from 2020-03-03 to 2020-04-23. The daily cases were transformed into weekly cumulative cases to reflect the SARS-CoV-2 virus incubation period of 5.6 to 7.7 days at 95% CI [45]. Therefore, a 7-day monitoring period is proposed.
The raw dataset discussed in Section 3 does not allow distance metrics to be applied directly as well as addressing the research challenges. Therefore, preprocessing and transformation are prerequisite to constructing an analytical dataset that fits the distance metrics. The subsections below aim to provide a high-level process from the construction of analytical dataset to proposal of distance metric utilization. The final subsection highlight the proposal of a unified modelling approach to avoid bias in a recommendation when relying merely on one distance metric.
4.1 Construction of Analytical Dataset
A well-designed data structure for the analytical dataset can lead to success in addressing the research challenges. However, identifying the right variables for the analytical dataset is a non-trivial task. One of the difficulty is to identify the right points of interest (POI) and their categories. In this study, two epidemiologists had been consulted to identify POI categories that are likely to affect COVID-19 transmission. The categories are Agriculture-based trading companies, ATM, Bazaar, Convenient Shop, F&B, Government Health Center, Hardware shops, Higher Education Institution, Hypermarket, Islamic School, Medical Services, Mini Market, Pharmacy, Shop, Shopping Complex, Supermarket, Toll, Trading Food, Transport, 24-Hour Shores, Wet Market, and Whole Sale Market. In total, 22 categories were identified to be included as part of the analytical dataset (D22poi).
Apart from the subsetted POI dataset (D22poi), the analytical dataset, Danalytics, was formed by aggregating De, Dpop, Dres, and Dcase. The columns in Tab. 2 are formed by concatenating all the variables from the datasets. The final analytical dataset consists of 111 variables and captures mostly numerical values except for the columns “Residential Area” and “Start Date.” Tab. 2 shows the partial analytical dataset with each record representing a patient’s details. The first record in Tab. 2 for instance, there is a patient who lives in LocA and within 500 m there are 321 elderly people with age more than 60, six convenient stores, and three places of worship. The first case at LocA was reported on 1 March 2020. The second record in the table shows a patient who lives at LocB, which is having higher population density and more POI as compared to LocA. The first case reported at LocB was 15 March 2020, which is two weeks after LocA had its first case. In this study, place of worship is one of the crucial variable because in Selangor, a major COVID-19 outbreak happened after the Tabligh event ended in middle of March 2020 [4].
Table 2: Analytical dataset with partial features
4.2 Measuring the Similarity Between Sites
The construction of analytical dataset has led to the design of algorithms that utilizes the distance metrics to tackle the research questions. There are two algorithms designed, with the first identifying top-3 most similar locations to a location of interest. The second algorithm estimates the time interval and total cases for the next most probable locations with at least one COVID-19 case.
Algorithm 1 was proposed to tackle the first research challenge in which it predicts the next most prob-able location with at least a confirmed case. The input to the algorithm is the analytical dataset, Danalytics, the distance metrics where 
, and a location of interest L. As shown in the algorithm, the location of interest is measured against all locations in the analytical dataset to elicit its respective similarity scores (SLmetric). This can be performed via the function metric–-Similarity–-Loc(). The output of the algorithm is a dataset, 
, which contains locations with the three highest similarity scores for L.
The output of Algorithm 1, 
, is fed into Algorithm 2 to estimate the mean duration for L to observe its first case. This is performed through the function dayDiff-mean() which calculates the mean of time interval for L’s similar locations that had demonstrated at least one case. The algorithm also estimates of mean total cases through the function case-mean().
The three different distance metrics produce different distance scores and the top-3 locations. Relying solely on one distance metric for decision-making can be risky and bias. Therefore, this study proposed a unified approach to decision-making by calculating the mean value from the output of the three different distance metrics. The high-level process flow is shown in Fig. 1, while the detailed steps are shown in Algorithm 3.
Figure 1: Decision-making strategy
Algorithm 3 discusses the procedure to aggregate the scores are given by different distance-based similarity measure techniques. To normalize the bias inherent within the metrics, inner-join was performed on the output of Algorithm 1 generated using different metrics. The aggregated dataset was then fed into two functions, namely calc-caseCount-mean() and calc-dayDiff-mean() to produce the final prediction on the total cases (
) and the day interval for that particular location of interest to have its first case (
).
5.1 Statistical Overview of Dataset
This section begins with a statistical overview of the dataset used in this study. Histograms are generated based on the following criteria: (i) sites (i.e., residential areas) and total cases, (ii) day interval between the first and last case of a site, (iii) day difference for start dates of two similar site, (iv) average similarity score between a site and its three most similar sites.
Fig. 2 shows that most of the residential areas in the Selangor state had experienced 5 cases and below, totalling up to 458 locations. There were 24 residential areas with cases between 5 and 10. There was only one location with more than 25 cases.
Figure 2: Residential areas with different number of cases
As shown in Fig. 3, there were 336 resident areas had their first and last cases within an interval of 5 days. From the raw dataset, most of the residential areas had either 1 or 2 cases only within 5 days period. Those residential areas that had more than 15 days interval contributed to 10.15% of the total residential areas.
Figure 3: Day intervals between the first and last case
The graph above shows the average distance scores between any one residential area and its top-3 similar ones (Fig. 4). When calculated via Euclidean distance, most of the residential areas depicted the distance less than 0.5, indicating that there are common characteristics among the residential areas in the Selangor state. To be specific, there are 77.1% of residential areas with average distance scores less than 0.5.
Figure 4: Average distance score between a residential area with its top-3 similar areas
Fig. 5 shows the distribution of duration between the start dates of two residential areas. It was found that in average the time interval between any two similar residential areas was 12 days. In other words, for any two similar residential areas, the latter residential area will likely to observe a case after a case was observed a area similar to it. However, low possibility could be observed for any two similar residential area to have a large time interval. This is shown in the figure that only one location with a time interval of more than 60 days.
Figure 5: Average duration between 2 start dates
5.2 Performance of Distance Metrics
5.2.1 Prediction of Total Cases
In this study, a total of 532 records were used for evaluation of the performance of the three distance metrics. These records were selected because they appeared in 
, 
, and 
. The predicted total cases were benchmarked against the actual total cases, and the evaluation metrics are Mean Absolute Error (MAE), Mean Square Error (MSE), and Root Mean Square Error (RMSE). As shown in Tab. 3, the results showed that Euclidean outperformed Bray Curtis and City Block in predicting the total case using MAE, however, the Unified Modeling approach outperformed the rest when MSE and RMSE were used.
Table 3: Total case prediction by three difference evaluation metrics and proposed a unified decision-making approach
5.2.2 Outbreak Duration Prediction
Tab. 4 shows the prediction of duration between the first and last case in a particular residential area.
Table 4: Prediction of outbreak duration
Comparing among the three distance measure techniques, Euclidean outperformed Bray Curtis and City Block for all three evaluation metrics. Comparing to the prediction of total cases, the Unified Modeling approach consistently depicted the lowest error rates in the prediction of duration between the first and last case when MAE, MSE, and RMSE was used.
5.2.3 Day Interval Between Start Dates of Two Similar Sites
One of the challenges in this research work was to predict the duration between the start dates of two similar residential areas. This would allow medical resources to be planned and optimized in the event that a COVID-19 patient is detected at a particular residential area. The proposed algorithms were able to predict the day intervals between a former and a latter residential area. Based on the experiment findings shown in Tab. 5, the Unified Decision-Making approach, had again, outperformed Bray Curtis, Euclidean, and City Block with 5.3842, 43.4475, and 6.5914, respectively.
Table 5: Difference in two start dates
Although there has not been a conclusive study of the SARS-CoV-2 virus that is causing the COVID19 pandemic now, it has generally been characterized by having high infectious rate and having varying incubation period. More worrying is when there are patients that are reported to be asymptomatic. Such characteristics have together explained the rapid development of an epidemic in Malaysia and worldwide. While waiting for the vaccine to be developed, one of the solutions to minimize outbreak is via active monitoring and mass screening. Performing mass screening can be taxing in both human and medical resources. A possible solution to the challenge is by deploying geospatial analytics. The research work presented in this paper attempt to answer the following questions: (i) how to identify the site (residential areas) where COVID-19 cases may appear next? (ii) what is the estimate of COVID-19 cases for that new residential area? (iii) what is the time interval before a COVID-19 case is observed at a similar residential area? Answering the questions require relevant datasets. Four different datasets were identified, and they are the patient addresses, population density, residential areas, daily cases, and point of interest. These datasets were transformed into an analytical dataset before feeding into distance-based similarity measure metrics.
In this study, three distance-based similarity measures had been employed in this study to measure the distance (a.k.a similarity) between two records. The distance metrics used are Euclidean distance, City Block distance, and Bray Curtis distance. Rather than relying on the results elicited by one distance metric, an unified modelling approach, which takes the average of 3 output, was proposed. The empirical findings suggested that the unified modelling approach is a promising approach to tackle the challenges listed above. The proposed approach outperformed any single distance metric such as Euclidean, City Block, and Bray Curtis in almost all the evaluation metrics, which have included the MAE, MSE, and RMSE. It is therefore suggesting that the Unified Decision-Making approach to similarity measures can be a solution to effective monitoring and proactive intervention.
The work can be extended to three perspectives. First is to deploy other types of distance metrics such as cosine distance, average distance, and weighted Euclidean. Second, the study could deploy machine learning approach such as classification and clustering to predict the total case and duration of an outbreak at a particular zone. Third, this research work has listed 22 categories of Point of Interest relevant to COVID-19 transmission. Further work could be done to rank the categories according to their importance towards influencing COVID-19 transmission.
Acknowledgement: The authors thank the Government of Selangor to provide partial COVID-19 patient information for the purpose of academic use.
Funding Statement: The authors received no specific funding for this study.
Conflicts of Interest: The authors declare that they have no conflicts of interest to report regarding the present study.
References
1. WHO. (2020). “Novel coronavirus (2019-ncov) situation reports,” . [Online]. Available: https://www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports.
2. Reuters. (2020). “Malaysia confirms first cases of coronavirus infection,” . [Online]. Available: https://www.reuters.com/article/china-health-malaysia/malaysia-confirms-first-cases-of-coronavirus-infection-idUSL4N29U03A.
3. Theedgemarketscom. (2020). “Three covid-19 cases confirmed in malaysia today, bringing total to 22,” . [Online]. Available: https://www.theedgemarkets.com/article/ two-new-confirmed-covid19-cases-malaysia.
4. T. Sukumaran. (2020). “How the coronavirus spread at malaysia’s tabligh islamic gathering,” . [Online]. Available: https://www.scmp.com/week-asia/explained/article/3075968/how-coronavirus-spread-malaysias-tabligh-islamic-gathering.
5. F. M. Today. (2020). “Mco extended 2 weeks to may 12,” . [Online]. Available: https://www.freemalaysiatoday.com/category/nation/2020/04/23/mco-extended-2-weeks-to-may-12/.
6. J. C. McGinty. (2020). “What will it take to flatten the coronavirus curve?,” . [Online]. Available: https://www.wsj.com/articles/what-does-it-mean-to-flatten-the-curve-to-fight-coronavirus-11585301402.
7. K. L. Ailawadi and P. W. Farris. (2017). “Managing multi- and omni-channel distribution: Metrics and research directions,” Journal of Retailing, vol. 93, no. 1, pp. 120–135.
8. E. T. Bradlow, M. Gangwar, P. Kopalle and S. Voleti. (2017). “The role of big data and predictive analytics in retailing,” Journal of Retailing, vol. 93, no. 1, pp. 79–95.
9. M. G. Dekimpe. (2020). “Retailing and retailing research in the age of big data analytics,” International Journal of Research in Marketing, vol. 37, no. 1, pp. 3–14.
10. D. Grewal, A. L. Roggeveen and J. Nordfalt. (2017). “The future of retailing,” Journal of Retailing, vol. 93, no. 1, pp. 1–6. [Google Scholar]
11. Y. Liu, H. Wang, G. Li, J. Gao, H. Hu et al. (2016). “ELAN: An efficient location-aware analytics system,” Big Data Research, vol. 5, pp. 16–21. [Google Scholar]
12. A. Trivedi and A. Singh. (2017). “A hybrid multi-objective decision model for emergency shelter location-relocation projects using fuzzy analytic hierarchy process and goal programming approach,” International Journal of Project Management, vol. 35, no. 5, pp. 827– 840. [Google Scholar]
13. M. Lukas and E. López-Morales. (2018). “Real estate production, geographies of mobility and spatial contestation: A two-case study in santiago de chile,” Journal of Transport Geography, vol. 67, pp. 92–101. [Google Scholar]
14. J. Wang, C.-H. Tsai and P.-C. Lin. (2016). “Applying spatial-temporal analysis and retail location theory to public bikes site selection in taipei,” Transportation Research Part A: Policy and Practice, vol. 94, pp. 45–61. [Google Scholar]
15. P. Cui, D. Ge and A. Gao. (2017). “Optimal landing site selection based on safety index during planetary descent,” Acta Astronautica, vol. 132, pp. 326–336. [Google Scholar]
16. M. Chavez, P. Berentsen and A. O. Lansink. (2012). “Assessment of criteria and farming activities for tobacco diversification using the analytical hierarchical process (ahp) technique,” Agricultural Systems, vol. 111, pp. 53–62. [Google Scholar]
17. M. Shaheen and M. Z. Khan. (2016). “A method of data mining for selection of site for wind turbines,” Renewable and Sustainable Energy Reviews, vol. 55, pp. 1225–1233. [Google Scholar]
18. M. Vasileiou, E. Loukogeorgaki and D. G. Vagiona. (2017). “GIS-based multi-criteria decision analysis for site selection of hybrid offshore wind and wave energy systems in Greece,” Renewable and Sustainable Energy Reviews, vol. 73, pp. 745–757. [Google Scholar]
19. J. Garcia, A. Alvarado, J. Blanco, E. Jimenez, A. Maldonado et al. (2014). “Multi-attribute evaluation and selection of sites for agricultural product warehouses based on an analytic hierarchy process,” Computers and Electronics in Agriculture, vol. 100, pp. 60–69.
20. T. Sahin, S. Ocak and M. Top. (2019). “Analytic hierarchy process for hospital site selection,” Health Policy and Technology, vol. 8, no. 1, pp. 42–50.
21. A. A. Merrouni, F. E. Elalaoui, A. Mezrhab, A. Mezrhab and A. Ghennioui. (2018). “Large scale pv sites selection by combining gis and analytical hierarchy process. Case study: Eastern morocco,” Renewable Energy, vol. 119, pp. 863–873.
22. M. Fabjanowicz, M. Bystrzanowska, J. Namiesnik, M. Tobiszewski and J. Plotka-Wasylka. (2018). “An analytical hierarchy process for selection of the optimal procedure for resveratrol determination in wine samples,” Microchemical Journal, vol. 142, pp. 126–134. [Google Scholar]
23. C. Eastwood, D. Chapman and M. Paine. (2012). “Networks of practice for co-construction of agricultural decision support systems: Case studies of precision dairy farms in Australia,” Agricultural Systems, vol. 108, pp. 10–18. [Google Scholar]
24. A. Y. Topraklı, A. Adem and M. Dagdeviren. (2016). “A courthouse site selection method using hesitant fuzzy linguistic term set: A case study for turkey,” Procedia Computer Science, vol. 102, pp. 603–610. [Google Scholar]
25. C. Zhou, F. Su, T. Pei, A. Zhang, Y. Du et al. (2020). “Covid-19: Challenges to GIS with big data,” Geography and Sustainability, vol. 1, no. 1, pp. 77–87. [Google Scholar]
26. Z. Arab-Mazar, R. Sah, A. A. Rabaan, K. Dhama and A. J. Rodriguez-Morales. (2020). “Mapping the incidence of the covid-19 hotspot in iran implications for travellers,” Travel Medicine and Infectious Disease, vol. 34, pp. 101630. [Google Scholar]
27. M. Lewin. (2020). “How better location intelligence can help businesses cope with covid-19_2020,” . [Online]. Available: https://resources.esri.ca/news-and-updates/how-better-location-intelligence-can-help-businesses-cope-with-covid-19. [Google Scholar]
28. A. Mollalo, B. Vahedi and K. M. Rivera. (2020). “GIS-based spatial modeling of covid-19 incidence rate in the continental united states,” Science of the Total Environment, vol. 728, pp. 138884. [Google Scholar]
29. F. G. Ashby and D. M. Ennis. (2020). “Similarity measures,” . [Online]. Available: http://www.scholarpedia.org/. [Google Scholar]
30. J. R. Bray and J. T. Curtis. (1957). “An ordination of the upland forest communities of southern wisconsin,” Ecological Monographs, vol. 27, no. 4, pp. 325–349. [Google Scholar]
31. J. Dattorro. (2005). Convex optimization & euclidean distance geometry. USA: Meboo Publishing. [Google Scholar]
32. J. Gower. (1982). “Euclidean distance geometry,” Mathematical Scientist, vol. 7, no. 1, pp. 1–14. [Google Scholar]
33. J. Han, M. Kamber and J. Pei. (2012). Data mining concepts and techniques. 3 ed., Waltham, Mass: Morgan Kaufmann Publishers. [Google Scholar]
34. S.-H. Cha. (2017). “Comprehensive survey on distance/similarity measures between probability density functions,” City, vol. 1, no. 2, pp. 1. [Google Scholar]
35. A. S. Shirkhorshidi, S. Aghabozorgi and T. Y. Wah. (2015). “A comparison study on similarity and dissimilarity measures in clustering continuous data,” PLoS ONE, vol. 10, no. 12, pp. e0144059. [Google Scholar]
36. A. K. Jain, M. N. Murty and P. J. Flynn. (1999). “Data clustering: A review,” ACM Computing Surveys, vol. 31, no. 3, pp. 264–323. [Google Scholar]
37. J. Mao and A. K. Jain. (1996). “A self-organizing network for hyperellipsoidal clustering (hec),” IEEE Transactions on Neural Networks, vol. 7, no. 1, pp. 16–29. [Google Scholar]
38. R. V. Rao, D. Singh, F. Bleicher and C. Dorn. (2012). “Weighted euclidean distance-based approach as a multiple attribute decision making method for manufacturing situations,” International Journal of Multicriteria Decision Making, vol. 2, no. 3, pp. 225–240. [Google Scholar]
39. D. Hand, P. Smyth and H. Mannila. (2001). Principles of Data Mining. Cambridge, Massachusetts, USA: MIT Press. [Google Scholar]
40. S. Government. (2020). “Dasar keselamatan ict,” . [Online]. Available: https://www.selangor.gov.my/. [Google Scholar]
41. HDX. Malaysia. (2020). “High resolution population density maps+demographic estimates,” . [Online]. Available: https://data.humdata.org/dataset/malaysia-high-resolution-population-density-maps-demographic-estimates. [Google Scholar]
42. JPPH. (2020). “Valuation and property services department Malaysia,” . [Online]. Available: https://www.jpph.gov. my/v3/ms/. [Google Scholar]
43. Brickz, Brickz. (2020). “True property prices,” . [Online]. Available: https://www.brickz.my/. [Google Scholar]
44. T. Malaysia. (2020). “Telekom malaysia,” . [Online]. Available: https://www.tm.com.my/Pages/Home.aspx, Accessed: 2020-04-27. [Google Scholar]
45. S. A. Lauer, K. H. Grantz, Q. Bi, F. K. Jones, Q. Zheng et al. (2020). “The incubation period of coronavirus disease 2019 (covid-19) from publicly reported confirmed cases: Estimation and application,” Annals of Internal Medicine, vol. 172, no. 9, pp. 577–582. [Google Scholar]
 | This work is licensed under a Creative Commons Attribution 4.0 International License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. |