Advanced Spatio-Temporal Neural Networks for Malaria Prevalence Forecasting in Sub-Saharan Africa

Accurate malaria prevalence forecasting is critical for timely interventions in Sub-Saharan Africa, where the disease remains a major health burden. Traditional regression models often struggle to capture the complex interactions between the epidemiological, environmental, and spatial factors that drive malaria transmission. This study develops and compares three distinct deep learning approaches for predicting the Plasmodium falciparum parasite rate in children (PfPR2-10), using historical prevalence data from 1900 to 2015 and climate variables such as temperature and precipitation sourced from NASAs POWER project. The first framework is a Spatiotemporal Transformer developed to capture intricate local dependencies between covariates. The second is a Patch-based Time Series Transformer (PatchTST) designed for robust point forecasting across temporal dimensions. The third approach employs a Spatio-Temporal Neural Ordinary Differential Equation (Neural ODE) model that learns the underlying continuous-time dynamics governing malaria prevalence, enabling smooth temporal interpolation and extrapolation. All three models were further extended with uncertainty estimation using Monte Carlo Dropout, providing probabilistic forecasts in addition to point predictions. Our findings highlight a key architectural trade-off: for this dataset, characterized by short time series and strong cross-feature interactions, the Spatiotemporal Transformer yielded the highest predictive accuracy. Conversely, while the Neural ODE demonstrated slightly lower point-forecast accuracy, it effectively captured temporal continuity and uncertainty dynamics, offering deeper insights into malarias spatio-temporal evolution. This comparative analysis provides a nuanced perspective, offering decision-makers a choice between highly accurate models for direct forecasting and probabilistic frameworks for risk assessment and strategic planning.