Spatiotemporal Machine Learning Models for Schistosomiasis Risk Mapping
Prosper Chidi Nwachukwu
Department of Biological Science, Clifford University, Owerrinta, Abia State, Nigeria.
Godson Chetachi Uzoaru *
Department of Computer Science, Clifford University, Owerrinta, Abia State, Nigeria.
*Author to whom correspondence should be addressed.
Abstract
Schistosomiasis remains a major neglected tropical disease driven by environmental, climatic, and socioeconomic determinants. Traditional surveillance approaches often rely on periodic field surveys, which may fail to capture fine-scale spatial and temporal transmission dynamics. The transmission dynamics of schistosomiasis are profoundly influenced by environmental and climatic conditions. The parasite’s life cycle is intricately dependent on specific freshwater snail species that serve as intermediate hosts, and which proliferate under suitable ecological circumstances. Factors such as temperature, rainfall patterns, water availability, and habitat stability play a critical role in determining snail distribution, abundance, and infection rates. Consequently, variations in these environmental parameters can significantly affect the intensity and spatial distribution of schistosomiasis transmission, underscoring the importance of ecological context in disease epidemiology. This study developed and evaluated spatiotemporal machine learning models to predict schistosomiasis risk using georeferenced epidemiological data integrated with environmental and climatic variables. A dataset comprising 2,500 location-specific infection records from 2015 to 2024 was analyzed alongside satellite-derived data, including rainfall, land surface temperature, normalized difference vegetation index (NDVI), elevation, and proximity to freshwater bodies. Random Forest (RF), Gradient Boosting (GB), and Logistic Regression (LR) models were implemented and compared. The Gradient Boosting model achieved the highest predictive performance (AUC = 0.91), outperforming RF (AUC = 0.88) and LR (AUC = 0.79). Risk mapping identified persistent transmission hotspots in low-altitude, high-rainfall zones with dense vegetation near water bodies. These findings demonstrate that spatiotemporal machine learning models can enhance targeted intervention strategies and optimise resource allocation in endemic regions.
Keywords: Schistosomiasis, risk mapping, spatiotemporal modeling, machine learning, GIS, neglected tropical diseases