Daily Sediment Yield Prediction Using Hybrid Machine Learning Approach

Abstract

In this study, four different soft computing AI techniques were tested for the prediction of sediment yield based on hydro-meteorological variables at Jondhara station, Seonath stream in Rajnandgaon district, India. In order to fulfill this purpose, the models namely, multilayer perceptron (MLP), support vector machine (SVM), multilayer perceptron coupled with genetic algorithm (MLP-GA), and support vector machine coupled with genetic algorithm (SVM-GA) models were employed. To select the optimal input variables, a statistical method such as the Gamma test was considered among several methods. Based on the results of the analysis, all models were evaluated by using the following statistical indices: Coefficient of Correlation (CC), room mean square error (RMSE) and percent bias (PBAIS). Overall, the performance of the studied models indicates that all of them are capable of simulation sediments yield at Jondhara station, Seonath river basin in a satisfactory manner. Comparison of results showed that the MLP-GA with CC = 0.988, RMSE = 0.006 and PBIAS = 0.000 in training period and CC= 0.990, RMSE = 0.007 and PBIAS = 0.000 in testing period for S-6 model and CC = 0.986, RMSE = 0.025 and PBIAS = -0.001 in training period and CC = 0.988, RMSE = 0.029 and PBIAS = -0.001 in testing period for S-13 model were able to yield better results than the other models considered. Furthermore, an SVM model is also observed to have some advantages over MLP models and SVM-GA models since it can represent the output data in a continuous manner by fitting a linear regression function to the output data, which has the advantage of making the model more precise than MLP and SVM-GA models.