Geospatial Based Surveillance of Malaria Risk in Dar es Salaam Using a Hybrid 3DCNN+LSTM and CA Model

Malaria remains as a significant public health burden in tropical and subtropical regions facing challenges in efficiently identifying and predicting risk areas. Conventional field‑survey methods used for mapping of Anopheles breeding sites are often costly, time‑consuming, and spatially incomplete. Therefore, therefore is a pressing need for a geospatially integrated surveillance framework for accurately mapping malaria risk and forecasting future risk dynamics to support targeted control efforts. A geospatial hybrid-modeling framework developed by integrating multi‑source remote sensing, Malaria and Climate datasets. Random Forest model employed solely to determine the relative importance of input variables, subsequently weighted and selected for inclusion in a deep learning architecture. The predictive model combined 3D Convolution Neural Network for capturing spatial patterns with a Long-Short Term Memory to learn temporal dynamics. The model trained against a baseline mean squared error (MSE) of 0.1. To improve spatial realism in the final risk maps, a Cellular Automata (CA) model incorporated using a 3×3 Moore neighborhood structure, with parameters calibrated at γ = 0.293 and β = 43.9 to enhance the spatial propagation of risks across neighboring cells. The framework successfully mapped malaria risks in Dar es Salaam with a Spearman correlation of 0.92 while Kigamboni South, Tundwi, and Msongola identified as high-risk areas. The hybrid 3DCNN–LSTM model prediction performance reduced training and validation losses by 97.3% and 90.2% respectively from the baseline with test MSE of 0.005. Prediction to 2060 exhibited a steady annual spatial increase in malaria risk of approximately 0.0022 risk units (slope = 0.077), demonstrating the model's ability for future risk estimation.