Employing Machine Learning Techniques for the Malaria Epidemic Prediction in Ethiopia

Malaria is one of the leading causes for death in Ethiopia. Even though currently its transmission decreases with many control efforts, the complexity of the problems is still very severe. Irregular epidemics have high consequences on society in term of morbidity and mortality. Government authorities are also incurring huge cost to control or eliminate the epidemic of malaria. It also costs the country in terms reduced productivity and increased school absenteeism. Accurate and reliable prediction of malarial epidemics is necessary for the health authorities to take the appropriate action for the control of the epidemic. Ethiopia uses a surveillance system to collect incidence and conventional analysis to predict and give an early warning system. However, there is a need of advanced suitable techniques to predict future epidemics.In this paper, we presented a framework which employs machine learning for the Malaria Epidemic Prediction (MEP) in Ethiopia based on the amount of rainfall, relative humidity, mean temperature, elevation and lag malaria cases. The machine learning techniques employed a more accurate opaque box model via a Support Vector Regression (SVR) and a transparent box model via an Adaptive Neuro Fuzzy inference System (ANFIS) to predict the malaria epidemic up to three months ahead. Thus, this framework allows us to gain a relatively high accuracy of prediction besides having transparency allowing us to understand the reasoning behind any prediction. The models were trained, validated and tested using 5 years (2013-2017) of historical climatic, elevation and malaria data from Ethiopia.