Metaheuristic-driven CatBoost model for accurate seepage loss prediction in lined canals

Abstract Precise estimation of seepage loss in lined canals is vital for improving water resource management, especially in water-scarce regions. This research focuses on optimizing Categorical Boosting (CatBoost) model hyperparameters using three advanced metaheuristic algorithms: Phasor Particle Swarm Optimization (PPSO), Dwarf Mongoose Optimization (DMO), and Atom Search Optimization (ASO). The objective was to develop high-accuracy hybrid models for predicting seepage loss, expressed as a dimensionless ratio, using multiple key canal design and liner characteristics as input variables. Six-hundred numerical datasets were gathered and split into 70% and 30% for training and testing stages, respectively. Rigorous analyses, including uncertainty evaluations and both visual and quantitative validation methods, were applied to assess the models' accuracy and effectiveness. Results demonstrated that the ASO-CatBoost model outperformed the standalone CatBoost model, achieving an R 2 of 0.993 and an RMSE of 0.321 in the testing phase. The ASO-CatBoost model demonstrated greater predictive accuracy, robustness, and generalization capabilities than both the PPSO-CatBoost and DMO-CatBoost models. Uncertainty analysis indicated that ASO-CatBoost exhibited the lowest uncertainty during both training and testing phases, highlighting its exceptional stability. Through SHapley Additive exPlanations (SHAP) and Partial Dependence Plots (PDP) analyses, liner hydraulic conductivity was identified as the most significant factor influencing seepage loss. Additionally, an interactive Graphical User Interface (GUI) was created to streamline the prediction process, offering engineers a user-friendly and effective tool for evaluating seepage loss. Graphical abstract