2026/01/01 by Dupsy Akoma, Yusuf Abdulakeem, Ganiyat Eshikhena +13 · 1 voice
Medicine · Environmental Science · Mathematics · #Malaria Research and Control #Hydrological Forecasting Using AI #Statistical Methods in Epidemiology
paper · doi:10.1016/j.dcit.2026.100084
openalex publication_date 2026/01/01 · openalex created_date 2026/05/31 · openalex updated_date 2026/06/19
Background Malaria transmission in Taraba State, Nigeria, remains a persistent public health challenge, shaped by climatic and ecological factors. Although the influence of weather patterns on malaria incidence has been widely studied, limited research has incorporated ecological zoning into predictive modelling frameworks. Methods We analysed malaria incidence data (2021–2023) from the District Health Information Software (DHIS2), alongside rainfall and temperature data from the Nigerian Meteorological Agency (NiMet). Data were cleaned, standardised, and encoded to include ecological and local government area (LGA) variables. Supervised machine learning models, including Decision Tree, Random Forest, and Extreme Gradient Boosting (XGBoost), were trained and validated using four-fold cross-validation. Model performance was assessed with mean absolute error (MAE), mean squared error (MSE), and root mean squared error (RMSE). Shapley Additive Explanations (SHAP) were applied to determine feature importance. Exploratory analyses employed one-way ANOVA to compare malaria incidence across ecological zones and Pearson correlation to test associations with climatic variables. Results Exploratory analysis revealed significant heterogeneity in malaria incidence across ecological zones (ANOVA: F = 5.21, p < 0.01), with the Montane Forest reporting the highest burden, largely driven by Sardauna and Bali LGAs. Rainfall showed a positive correlation with malaria incidence ( r = 0.42, p < 0.01), while temperature was negatively correlated ( r = −0.36, p < 0.05). Across all models, ecological zone consistently emerged as the strongest predictor of malaria incidence, outweighing climatic factors. Random Forest achieved the best predictive accuracy (MAE = 452.6, RMSE = 675.5). SHAP analysis confirmed that geographic variables (zone and LGA) exerted the greatest influence on malaria predictions. Conclusion Machine learning approaches may provide a useful framework for exploring malaria transmission dynamics in climate-sensitive and ecologically diverse settings. The findings of this study suggest that ecological zones may be important determinants of malaria incidence in Taraba State. These results indicate that incorporating geographic heterogeneity into malaria surveillance and control strategies could help support more data-driven and locally tailored interventions