Comparing Classification Models for Predicting Malaria: A Case Study of Malaria Incidence in Kenya

Accurate prediction of malaria incidence is indispensable in helping policy makers and decision makers intervene before the onset of an outbreak and potentially save lives. Various classification models have been fitted in prior studies to describe both climatic and non-climatic factors that influence the spread of malaria. However, there have been no comprehensive studies comparing classifier algorithms to find which best describes the spread of malaria. This study fits five machine learning models to real malaria data for Kenya in a bid to establish which model has the highest predictive capability in the presence of various climatic and non-climatic predictor variables. Methods described in the available literature to forecast malaria involve complex numerical simulations for malaria transmission. This study presents a data-driven binary classification approach for the prediction of malaria incidence. Various metrics based on the confusion matrix are computed and used as a basis for comparing the fitted models. The results infer that Random Forest is the best model with 76% accuracy, 64.67% precision, 40.75% recall, and 90.71% specificity.