Determining Seepage Loss Predictions in Lined Canals Through Optimizing Advanced Gradient Boosting Techniques

Ensuring accurate estimation of seepage loss is critical for advancing water sustainability, especially in water-scarce regions. This study is aimed at evaluating the performance of three gradient boosting models: Xtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and Categorical Gradient Boosting (CatBoost) in predicting seepage loss in lined canals. A dataset comprising 600 samples was collected and split with a 70:30 split for training and testing stages. Four dimensionless inputs were considered: canal bed width to water depth ratio (b/y), side slope (z), liner to soil hydraulic conductivity ratio (k_L/k), and liner thickness to water depth ratio (t_L/y), while the seepage loss per unit canal length to the product of soil hydraulic conductivity and water depth (q/ky) was the output. The models’ hyperparameters were optimized using the Bayesian Optimization (BO) technique, with five-fold cross-validation ensuring rigorous evaluation. Comprehensive analyses, including uncertainty assessments, visual and quantitative methods, and Akaike Information Criterion (AIC), were employed to validate model effectiveness. CatBoost consistently outperformed the other models, achieving the highest R² (0.998) and lowest RMSE (0.189), highlighting its ability to handle complex data patterns and its superior optimization process. The study also incorporated SHapley Additive exPlanations (SHAP) and Partial Dependence Plot (PDP) analyses, revealing that the liner’s hydraulic conductivity had the most significant impact on seepage loss predictions. The findings emphasize the importance of proper hyperparameter tuning and demonstrate the CatBoost model’s robustness and reliability. An interactive graphical user interface (GUI) was developed to facilitate practical applications, enabling engineers to predict seepage loss quickly and economically.