Spatial prediction of human brucellosis susceptibility using an explainable optimized adaptive neuro fuzzy inference system.

Abstract

Brucellosis, a zoonotic disease caused by Brucella bacteria, poses significant risks to human, livestock, and wildlife health, alongside economic losses from livestock morbidity and mortality. This study improves Human Brucellosis Susceptibility Mapping (HBSM) by integrating the Adaptive Neuro-Fuzzy Inference System (ANFIS) with meta-heuristic algorithms, including Genetic Algorithm (GA) and Particle Swarm Optimization (PSO). Boruta-XGBoost identified key covariates, while VIF and tolerance tests addressed collinearity, and Shapley additive explanation (SHAP) values enhanced model interpretability. In Mazandaran province, Iran (2012-2018), the hybrid ANFIS-PSO model demonstrated superior performance (RMSE: 0.5076; R²: 0.6980). SHAP analysis highlighted mean elevation, NDVI, and relative humidity as the most impactful covariates, while max evaporation and precipitation had minimal influence. ANFIS-based models outperformed Support Vector Regression (SVR), offering a robust framework for brucellosis control. This approach enables effective interventions and resource allocation, with potential for improvement through advanced algorithms and greater interpretability.