Evaluation of advanced AI models for accurate prediction and uncertainty analysis of aeration inception in hydrofoil-crested stepped spillways.

Accurate prediction of the aeration inception point (Li) in hydrofoil-crested stepped spillways (HSSs) is crucial for preventing cavitation damage and ensuring safe energy dissipation and flow regulation. This study aims to develop an advanced machine-learning (ML) framework for reliably estimating Li and quantifying its predictive uncertainty using a unique experimental database. A comprehensive set of laboratory measurements from sixty HSS configurations was compiled and transformed into a dimensionless form based on hydraulic similarity principles. After outlier removal using Isolation Forest and normalization with StandardScaler, the dataset was split into training (75%) and testing (25%) subsets using a fixed random seed. A 5-fold cross-validation scheme was then applied during training to improve model robustness. Five state-of-the-art AI models for tabular data—TabPFN, HistGBoost, M5Prime, AutoInt, and SAINT—were optimized using Multi-Verse Optimizer (MVO) and Particle Swarm Optimization (PSO). Model performance was evaluated using conventional error metrics, Taylor diagrams, and NDGF analysis. SHAP and Explainable Boosting Machine (EBM) were used to interpret the input–output relationships. The results demonstrate excellent predictive skill, with TabPFN-PSO achieving the best generalization on the test set and HistGBoost-MVO exhibiting the most stable behavior under uncertainty analysis. Feature-importance analysis revealed the roughness Froude number as the dominant control on Li. The novelty of this work lies in integrating modern ML models, metaheuristic hyperparameter tuning, and explainable AI within a hydraulically informed framework. This integration enables accurate and interpretable prediction of aeration inception. These findings highlight the strong potential of ML-based modeling as a powerful decision-support tool for the design and optimization of stepped spillways.