Forecasting visceral leishmaniasis in Sudan using hybrid wavelet based deep learning models on climate driven...

Visceral leishmaniasis (VL) is a climate-driven disease with a complex epidemiological pattern and is difficult to predict, especially in regions like eastern Sudan, where disease patterns vary substantially from season to season. In this study, our objective was to forecast the monthly incidence of VL in Gedaref State based on key climatic drivers-precipitation, temperature, and humidity. To capture the complex dynamics of disease transmission, we propose a new hybrid model based on wavelet transforms and state-of-the-art deep learning models to decompose multi-scale patterns and learn linear and nonlinear relationships.. We developed and tested hybrid models: Wavelet-Gaussian Process Regression (Wavelet-GPR), Wavelet-Spatiotemporal Graph Neural Network (Wavelet-StemGNN), and Wavelet-Temporal Convolutional Network with Bidirectional LSTM (Wavelet-TCN-BiLSTM). These were compared against a traditional Vector Autoregressive model (VAR) as a baseline. We employed a fixed temporal split (2000-2018 for training; 2019-2022 for testing). All models were implemented using Python and their performance was evaluated using key forecasting metrics: RMSE, MAE, R², and MAPE. Our wavelet-GPR model outperformed both baseline and other deep learning methods in terms of prediction accuracy (RMSE = 43.93, MAE = 32.85, R² = 0.93, MAPE = 14.6%), with the smallest prediction errors and well-calibrated 95% predictive intervals. The results of our study underscore the possibility of improving the disease surveillance and early warning systems by means of hybrid wavelet-based artificial intelligence models, particularly in resource constraints situations. The projection may be employed for monthly early warning bulletins and the strategic pre-positioning of diagnostics and medicines in Gedaref. The method is straightforward to modify to different VL locations in Sudan and East Africa, and it requires much re-calibration. Future projects will include more layers of environmental and socio-economic data to make predictions and planning operations easier.