Performance Evaluation of LS-SVR Model in Predicting Scour Depth in Bridge Piers

Abstract

In this research work, two machine learning models including Least Squares Support Vector Machines (LS-SVR) and Adaptive Neuro-Fuzzy Inference System (ANFIS) were used to predict the scour depth around the bridge piers. For this purpose, 240 data series including pier geometry, flow condition, sediment characteristics, and some dimensional parameters were used. Dimensional and no dimensional parameters were considered. The performance of the models was evaluated using root mean square error (RMSE), mean absolute percentage error (MAPE), and Nash–Sutcliffe efficiency (NSE) criteria. The results showed that in both models, the use of dimensional parameters for prediction leads to high prediction accuracy. The comparison between the models also showed that the LS-SVR algorithm with the criteria RMSE=46.84, MAPE=38.03, NSE=0.62 for the test data of the first model and RMSE=28.62, MAPE=38.97, NSE=0.67 for the test data results of the second pattern are more accurate than the ANFIS algorithm. This research indicates that machine learning models are a suitable alternative to empirical models in predicting scour depth of bridge piers.

Keywords

• anfis
• bridge piers
• ls-svr
• machine learning
• prediction
• scour depth