Resurgence Prediction of Ten Infectious Diseases under Surveillance in Senegal

In this paper, there are proposed two multi-class predictive models for estimating the resurgence probability of ten infectious diseases under epidemic surveillance in Senegal. The first model is a Multiple Binary Random Forest (MBRF), which utilizes the ranger function with Gini criterion and allows to separately predict each of the ten diseases while taking account of their interdependencies. The second model is a Multi-Output Decision Tree (MODT), which introduces an inertia criterion (calculated with Chi-square distance) as the node impurity measure and allows to simultaneously predict all of ten diseases. Data come from the global disease surveillance database of the Ministry of Health, and contain information, on 68698 instances, related to disease's, district's as well as patient's characteristics. The results showed that, during the study period (January 2018 to November 2022), these ten pathologies recorded an average resurgence probability of 12.2\%, except for Poliomyelitis, which had a lower score estimated at 2.4%, and Covid-19 which showed a fairly high resurgence rate hovering 60%. Compared to standard algorithms such as: multi-class random forests (MCRF) and multinomial logistic regression (MLR), our two models provided better performance. For example, for F1-score, we have: MBRF (0.9999), MODT (0.8572), MCRF (0.8451), MLR (0.8211)