Chapter 26 - Sediment transport with soft computing application for tropical rivers

Abstract

This research revised the existing sediment transport equation for rivers in Malaysia. The current equations of Ariffin (2004) and Sinnakaudan et al. (2006) were modified by using MLR and machine learning programs, namely Evolutionary Polynomial Regression (EPR), Multi-Gene Genetic Programming (MGGP), and M5 tree model (M5P). Among the three machine learning models, in terms of coefficient of determination (R2), Nash-Sutcliffe coefficient of Efficiency (NSE), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE), EPR were able to give the best prediction model in the evidence of Revised Ariffin (2004) model (R2 = 0.922, NSE = 0.913, RMSE = 3.305, MAE = 1.552), followed by MGGP (R2 = 0.787, NSE = 0.784, RMSE = 5.217, MAE = 3.054) and M5P (R2 = 0.786, NSE = 0.762, RMSE = 5.467, MAE = 1.561). The trend was also the same for Revised Sinnakaudan et al. (2006) whereby EPR had an excellent prediction accuracy model (R2 = 0.884, NSE = 0.848, RMSE = 4.377 ,MAE = 2.137), followed by MGGP (R2 = 0.787, NSE = 0.784, RMSE = 5.207, MAE = 3.054) and M5P (R2 = 0.622, NSE = 0.615, RMSE = 6.961, MAE = 1.994). In terms of Discrepancy Ratio (DR), only M5P of both Revised Ariffin (2004) (73.46%) and Revised Sinnakaudan (2006) (73.36%) produced better results than MLR (66.36%). However, the data did not distribute well and is rather flattening at the lower total bed material load rate. Machine learning is excellent at improving the prediction distribution at the high-value data but lacks accuracy compared to the observed value at the lower data value. This is mainly due to the type of regression algorithm used and sample size used in this study.