Traffic Flow Forecasting by Combination of SVM with PCA

Abstract

For improving traffic flow forecasting precision, a forecasting method that combines nonlinear regression Support Vector Machines (SVM) with Principal Component Analysis (PCA) was proposed. PCA was used to extract features from forecasting variables and produce fewer principal components. These principal components were input to nonlinear regress SVM for traffic flow forecasting. The kernel parameters of the SVM were determined with Bayesian inference. The efficiency of the method was illustrated through analyzing Jinan urban traffic flow data. Analysis results show that the traffic flow forecasting method that combines nonlinear regression SVM with PCA can not only improve forecasting precision but reduce computation cost, which can improve the real-time performance of forecasting. The forecasting precision of the proposed method is higher than that of commonly used traffic flow forecasting methods.