MalariVis: A Bi-Modal Machine Learning System for Malaria Case Forecasting and Outbreak Prediction

Abstract

Every 50 seconds, malaria claims another life, an annual cost of over 600,000 people. Eradication efforts have been widespread, yet cases have increased over time, and life-saving care often remains out of reach for those who need it most. While malaria is entirely treatable, hundreds of thousands continue to pass away each and every year; change is necessary. This study aimed to develop a new approach to the problem, creating a dual-focused system for both malaria incidence forecasting (regression) and outbreak prediction (classification), one to four months in advance. Built on clinical data from the National Institute for Pharmaceutical Research and Development and climatic datasets from the POWER Project, fourteen robust models were tested in Abuja, Nigeria: nine for regression and five for classification. The Seasonal Autoregressive Integrated Moving Average Model proved best for the first task, with an average R2 score of 0.92 across four months of forecasting and statistical significance (p<0.05). The Long Short-Term Memory Model performed just as well for the second task, with an average accuracy of 92.75% (p<0.05). Furthermore, in-depth feature analysis was conducted, showing wind speed, rainfall, and humidity as key driving factors of malaria incidence. The finalized models were implemented into MalariVis, a free, innovative, and easily-accessible application for malaria forecasting, enabling health officials and civilians alike to prepare for upcoming outbreaks before they happen, minimizing upcoming casualties. MalariVis could be the future of malaria prediction and prevention, saving lives on a pathway to eradication.