Hybrid AI accurately forecasts deadly tropical disease in Sudan

Visceral leishmaniasis (VL) is a severe, climate-driven disease with complex transmission patterns, making outbreaks difficult to predict in regions like eastern Sudan. To tackle this volatility, researchers focussed on Gedaref State, developing a hybrid modelling system that analyses climatic drivers such as precipitation, temperature, and humidity.

The team employed a novel technique combining wavelet transforms—which decompose complex data patterns—with advanced deep learning algorithms to interpret linear and nonlinear relationships. Specifically, the Wavelet-Gaussian Process Regression (Wavelet-GPR) model proved exceptional. When tested against data from 2019 to 2022, it outperformed traditional Vector Autoregressive baselines and other neural networks, achieving a prediction accuracy score (R²) of 0.93.

This high-precision modelling offers a vital tool for public health in resource-constrained situations. By accurately forecasting monthly incidence, officials can issue early warning bulletins and strategically pre-position diagnostics and medicines where they are needed most. The authors note that this method is readily adaptable to other VL-endemic locations across East Africa.