Scour depth prediction around bridge piers of various geometries using advanced machine learning and data augmentation techniques

Scour around bridge piers is a major concern for structural safety, yet traditional estimation methods often lack accuracy. This study explores advanced machine learning models to improve scour depth prediction for various pier shapes. We compare three models: Random Forest (RF), Support Vector Regression (SVR), and a Genetic Algorithm-optimized Multilayer Perceptron (GA-MLP). A comprehensive dataset, enhanced through data augmentation, was used to train the models. This dataset covers diverse flow conditions, pier geometries, and sediment characteristics. the results demonstrate that Random Forest (RF) achieved the highest predictive accuracy, with a Mean Squared Error (MSE) of 0.018, a Coefficient of Determination (R²) of 0.90, and a Kling-Gupta Efficiency (KGE) of 0.84 on the test dataset.