Comparison of Logistic Regression Models, Random Drills and Support Vector Machines (SVM) for the Epidemiological Surveillance of Ten (10) Infectious Diseases in Senegal

Objective: The objective of this study is to analyze data from epidemiological surveillance and take into account propagation factors to track and monitor 10 infectious diseases in Senegal which are: Covid-19, Measles, Poliomyelitis (AFP), Dengue, Meningitis, Rift Valley Fever (Rift), Crimean-Congo Hemorrhagic Fever (CCHF), Chikungunya (CHIK), West Nile virus fever (WN) and Yellow Fever (YF). Method : The methodology is based on a comparative analysis of the performances of three supervised learning methods: logistic regression, random forests and Support Vector Machines (SVM). Data visualization methods such as Multiple Correspondence Analysis (MCA) and similarity index (Cramer's v) are also used. The data analyzed covered the period from January 2018 to November 2022 and amounted to 113,847 suspected cases. Findings: The results show that the methods have almost similar performances with regard to the choice of the number of selected parameters (4 variables) but differ in terms of total classification error rates, with a rate of 2.73% lower for forests random, 3.10% for logistic regression and 3.32% for Support Vector Machines. Advanced age ([40, 60] years) is identified as a direct risk factor in relation to the frequency of distribution of confirmed cases of the 10 infectious diseases. Other indirect risk factors linked to the distribution of confirmed cases are also identified and are linked to the physical environment of health districts and climatic factors. The results highlighted the ability of the random forest method to analyze surveillance data and identify complex relationships between risk factors and the distribution of infectious diseases. This study contributes to improving infectious disease surveillance and shows the importance of integrating environmental factors, socioeconomic parameters, as well as behavioral and climatic factors into data and analyses. Novelty: The results obtained made it possible to propose an algorithm based on the HCPC method which combines the techniques of Principal Component Analysis (PCA) and those of Ascending Hierarchical Classification (CAH), with the application of the principle of paragons. Keywords: Epidemiological data, Health surveillance, Logistic regression, Random Forests, Support Vector Machine