Environmental and Climate-Resilient Approaches to Scour Depth Prediction: From Empirical Equations to Advanced Machine Learning Models

Scour around bridge structures is a major cause of bridge failure, disturbing sediments and aquatic habitats; therefore, accurate prediction for structural safety and river ecosystem protection is necessary. This study bridges the gap between research and practice by highlighting the disconnect between Machine Learning (ML)- based prediction and current conventional design methods, and systematically compares their performance using consistent statistical indicators. The analysis showed that dataset size ranged from 100 to 871 samples, with an inconsistent relationship between sample size and predictive accuracy (R² = 0.850–0.959), whereas a mid-sized dataset of 500 samples performed best (R² = 0.959), indicating that no single factor determines optimal model performance. Therefore, this study recommends applying large-scale field datasets alongside laboratory datasets in different machine learning models to achieve more precise and reliable results. Such advancements can reduce unnecessary streambed disturbance and improve prediction reliability, supporting more sustainable river infrastructure.