Explainable AI via PSO-optimized CatBoost and SHAP analysis: a case study on nanocomposite membrane performance

This study presents an explainable artificial intelligence (AI) framework for modeling complex nonlinear systems using a hybrid PSO–CatBoost approach enhanced with SHAP-based interpretability. Particle Swarm Optimization (PSO) was employed to automatically tune CatBoost hyperparameters, improving predictive accuracy and generalization in structured multidimensional datasets. Model robustness was validated using five-fold cross-validation, followed by evaluation on an independent hold-out test set. The framework was applied to predict the performance of amine-functionalized MWCNT nanocomposite membranes for simultaneous removal of organic pollutants, lead, and salts from wastewater. The optimized model achieved high predictive accuracy, with R² values of 0.93, 0.94, and 0.97 for the respective pollutants. Inference latency remained low, supporting practical deployment. SHAP analysis provided transparent interpretation of feature contributions, enhancing model reliability and physical consistency. The results demonstrate that integrating metaheuristic optimization, gradient boosting, and explainable AI offers a robust and scalable approach for modeling complex environmental systems.