Stream flow forecasting using principal component analysis and least square support vector machine

Abstract

This paper investigates the ability of Least Square Support Vector Machine (LSSVM) with Principal Component Analysis model as data preprocessing tool to improve the accuracy of stream flow forecasting. The objective of this study is to evaluate the potential of a Principal Component Analysis (PCA) method, by extracting the principal components from lagged input of monthly stream flow data. To assess the effectiveness of this model, monthly stream flow record data from two stations: Muda and Selangor River in Malaysia, have been used as the case study. The performance of the LSSVM model using PCA as data preprocessing tool is compare with single LSSVM model using various statistics measures. The comparison results indicate the LSSVM with PCA model is a useful tool and a promising new method for stream flow forecasting. The results showed that LSSVM with PCA as data preprocessing technique were found to provide a better representation and good forecasting results for both of the rivers.