Abstract

Accurate identification of flow regimes is crucial for understanding and analyzing flow behavior over stepped spillways. This study evaluates the ability of three machine learning regression models, namely AdaBoost Regressor (AdaBoost), Extra Trees Regressor (ETR), and Extreme Gradient Boosting (XGBoost), to predict three flow regimes (nappe flow, transition flow, and skimming flow) over stepped spillways. A dataset of 126 samples, including the ratio of critical flow depth to step height (h<inf>c</inf>/h), chute slope (α), and flow condition, was collected from a hydraulic experimental study. Two data splitting ratios were used for model training and testing: 110:16 and 84:42. The models were evaluated and compared using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), coefficient of determination (R²), and Overall Index (OI). The findings revealed that AdaBoost, ETR, and XGBoost achieved higher accuracy than previous studies that employed Artificial Neural Networks (ANNs) and the Adaptive Neuro-Fuzzy Inference System (ANFIS), with R2 and OI exceeding 0.90 for both data-splitting ratios. The AdaBoost model demonstrated the highest performance, followed by ETR and XGBoost, respectively. This study contributes to the advancement of knowledge of machine learning models, particularly in their application to hydraulic engineering contexts.