Comparison between three ML models to estimate the river’s total sediment loads
Comparison between three ML models to estimate the river’s total sediment loads
Seied Hosein Afzali1 Reza Piraei2 Majid Niazkar3
1) 1. Associate Professor of Civil Eng, Department of Civil and Environmental Engineering, Shiraz University, Shiraz, Iran. Corresponding Author,
2) 2. Ph.D. Student of Civil Engineering, Department of Civil and Environmental Engineering, Shiraz University, Shiraz, Iran.
3) 3. Ph.D., Faculty of Science and Technology, Free University of Bozen-Bolzano, Piazza Università 5, 39100 Bolzano, Italy:
IOI Code
:
XHHA-AZZDZC
Publication :
5th. International Congress On Engineering, Technology and Applied Sciences(etas2023.com)
Abstract :
Estimating total sediment loads in river management poses a significant challenge due to the costly and time-consuming nature of direct measurement. This study employs the eXtreme Gradient Boosting (XGBoost) model and compares its efficacy with two other machine learning (ML) models such as Artificial Neural Networks (ANN) and AdaBoost. Utilizing 543 data points from the United States Geological Survey, various models were trained and tested. Results indicate that XGBoost outperforms other methods across six performance metrics. Specifically, XGBoost yields root mean square errors of 216 and a determination coefficient of 0.95, surpassing ANN s metrics of 316.23 and 0.87, respectively. The study employs XGBoost feature importance and SHapley Additive exPlanations (SHAP) to interpret sediment predictions and assess the significance of each feature. Notably, the analysis reveals that water surface width predominantly influences XGBoost estimations (72%). SHAP analysis further corroborates the importance of water surface width in the final predictions. The findings suggest promising avenues for employing XGBoost in water resources management.
Keywords : total sediment load data driven models machine learning xgboost shap