Scour Depth Prediction Around a Circular Pier via ANN, SVM, KNN, and RF

Abstract

Ensuring bridge safety is vital to prevent major economic and societal impacts. Scour at bridge piers represents a major risk and is determined by factors such as approach flow depth, pier diameter, critical flow velocity, sediment properties, and pier shape. While several empirical equations have been proposed to predict scour depth under clear-water and live-bed conditions, they often show limited accuracy due to their basis in controlled laboratory experiments, narrow data ranges, and inability to capture nonlinear effects. To overcome these issues, this study utilizes machine learning models—including artificial neural networks (ANNs), support vector machines (SVMs), K-nearest neighbors (KNNs), and random forests (RFs)—to estimate normalized scour depth around single circular piers. Model performance is compared to conventional empirical equations using 544 laboratory observations split into training, validation, and testing sets. Accuracy is measured using metrics such as root mean square error (RMSE), mean absolute error (MAE), coefficient of determination (R²), and Nash–Sutcliffe efficiency (NSE). Visualization methods illustrate the agreement between ML predictions and traditional estimates. Sensitivity analysis shows the pier width?to?flow depth ratio (b/y) as the most critical parameter for predicting normalized scour depth (ds/y). KNN achieves the best results (RMSE = 0.0058, MAE = 0.0943, R² = 0.894, and NSE = 0.891). Overall, the findings demonstrate that ML-based approaches can improve scour depth predictions, supporting better bridge design and reducing failure risks from scour.