Leveraging Routine Data and Local Weather Variables for Malaria Case Prediction in Tanzania: A Deep Learning Approach

Malaria remains a life-threatening disease predominantly transmitted by Anopheles mosquitoes, with Sub-Saharan Africa bearing the brunt of its impact accounting for 95% of global cases. Tanzania, a Sub-Saharan nation, contributes 4.1% of global malaria deaths, ranking among the top four countries most burdened by the disease. Despite the deployment of effective interventions such as Long-Lasting Insecticidal Nets (LLINs) and Indoor Residual Spraying (IRS), malaria persists, particularly in certain ecological hotspots within the country. Tanzania's diverse ecological settings, ranging from wet highlands to dry lowlands, influence malaria prevalence and transmission dynamics. Weather conditions, including temperature and rainfall, significantly affect malaria transmission, as warmer and wetter conditions favor the growth of Anopheles mosquitoes. This study leveraged nine years of routine malaria case data from 25 districts alongside weather data from the NASA portal to develop machine learning-based prediction models capable of forecasting daily malaria cases at the district level for the next 16 days. Various algorithms, including XGBoost, Recurrent Neural Networks (RNNs), and K-Nearest Neighbors (KNNs), were trained and evaluated for their predictive performance. Among these, the XGBoost model outperformed others, demonstrating superior accuracy and robustness. Findings highlight humidity and temperature as the most significant variables influencing malaria incidence, underscoring the critical role of localized weather conditions. To translate these findings into practical applications, the study developed an R Shiny app that integrates the predictive model with forecasted weather data from the OpenWeatherMap API, providing stakeholders with district-level malaria forecasts to enable timely and targeted interventions. It is recommended that this predictive framework be scaled nationally, incorporating additional factors such as socioeconomic and intervention coverage data to further enhance its accuracy. Capacity building for district health managers and real-time satellite data integration are crucial for improving prediction reliability and supporting more effective malaria control efforts in Tanzania.