Time-Series Prediction of Streamflows of Malaysian Rivers Using Data-Driven Techniques

Abstract

A reliable and continuous streamflow simulation capability is essential for systematic management of water resource systems. Thus, predicting streamflow is important for water management and flood control. This study evaluated the effectiveness of a few data-driven procedures, such as the least squares support vector machine (LS-SVM), M5P tree, and random forest (RF) algorithm for estimating streamflows of the Bernam and Tualang rivers of Malaysia. Three standard statistical measures, i.e., correlation coefficient (CE), root mean square error (RMSE), and mean absolute error (MAE), were used to evaluate the performance of the developed model. The performance of RF-based models was found to be higher than that of LS-SVM and M5P-based models with respect to predicting streamflow for both the rivers.