Short-term traffic flow forecasting based on feature selection with mutual information

Abstract

Abstract. Traffic flow forecasting is related to many traffic variables, and how to select appropriate traffic variable combination is very important to traffic flow forecasting, which can reduce the cost of calculation and improve the forecasting precision. In this paper, a feature selection technique with mutual information is proposed for this purpose. Firstly, the mutual information is used to evaluate the relevance and redundancy of variables, and feature selection is used to select the relevant variables and filter out the redundancy between the selected variables. Secondly, BP neural network is used as the forecasting engine. Finally, a numerical example of traffic flow data from Pems is used to verify the forecasting performance of the proposed method, the results indicate that the proposed method can effectively reduce the cost of calculation and also improve the model forecasting precision.