Daily suspended sediment simulation using machine learning approach

• Suspended sediment is modelled using least square support vector regression (LSSVR) for Kopili River basin.
• LSSVR results are compared with six different soft computing methods.
• LSSVR and ANN models are superior to the other models for Kopili River basin.
• Sediment data can be successfully simulated by LSSVR using hydro-meteorological data of the study area.

In the present study, an attempt has been made to compare the performances of six different soft computing techniques for suspended sediment prediction using hydro-meteorological variables as input at Kopili river basin in India. For this purpose, the models namely, artificial neural network (ANN), radial basis function neural networks (RBFNN), least square support vector regression (LS-SVR), multi-linear regression (MLR) and decision tree models such as Classification and Regression Tree (CART) and M5 model tree were employed. The input variable selection was derived considering the statistical properties such as auto-, partial-, and cross-correlation function of the time series. The performances of all the models were judged using statistical indexes i.e. Coefficient of Correlation (CC), Nash–Sutcliffe coefficient (ENS) and RMSE-observations Standard deviation Ratio (RSR). Overall performance of the models shows that, all the studied models are able to simulate suspended sediment of the Kopili River basin satisfactorily. Comparison of results showed that the LSSVR (ENS = 89.00%) and ANN (ENS = 88.78%) models were able to produce better results than the other models investigated. This study also suggests that, an M5 model tree, which obtains a continuous representation of the output data by fitting a linear regression function to the data, has certain advantages over CART and MLR model because they offer more insight into the generated model.