Predicting Energy Loss Over Vegetated Dikes Utilizing Machine Learning Techniques

Abstract

Vegetated dike has a significant role in energy loss of flooding, however, it is a challenging task to accurately predict the energy loss. Therefore, the present research work attempted to estimate the energy loss of flood over a vegetated dike utilizing machine learning techniques (ML) including random forest (RF) and extreme gradient boosting with particle swarm optimization (XGBoost-PSO). Dataset of various parameters like Froude number (Fr), velocity reduction (Vo/V), ho/h, (ho: initial water depth, h: water in a flume with vegetated dike), ho/B, (B: channel width), and energy loss (E) was calculated from the experiment performed in a controlled laboratory setting. Moreover, SHAP analysis was performed to investigate the impact of critical parameters on energy loss. The result of the findings demonstrates the superior performance of the XGBoost-PSO due to a higher R-value of 0.99 and a lower MSE value of 0.0345. The SHAP analysis result indicates that in the case of the RF model parameter ho/h has a significant impact on energy loss while ho/B in the case of the XGBoost-PSO model. The findings of the present research provide a precise estimation of the energy loss while designing a vegetated dike in a flood-prone region.