Short-Term Traffic Flow Prediction Based on XGBoost

Abstract

Fast and accurate short-term traffic flow prediction is an important precondition for traffic analysis and control. Due to the fact that the short-term traffic flow has nonlinear characteristic and changes randomly, concurrent computation is difficult for traditional machine learning algorithms. In this paper, a traffic flow prediction model combining wavelets decomposition and reconstruction with the extreme gradient boosting (XGBoost) algorithm is proposed to predict the short-term traffic flow. First, in the training part, wavelet de-noising algorithm is utilized to obtain the high and low frequency information of target traffic flow. Secondly, the high frequency information of traffic flow is processed by threshold method. After that, the high and low frequency information is reconstituted as the training label. Finally, the de-noised target flow is sent to the XGBoost algorithm for training to predict traffic flow. In this way, the trend of the traffic flow in each sample period is retained, and the influence of the short-term high frequency noise is reduced. The proposed traffic flow prediction method is tested base on the traffic flow detector data collected in Beijing, and the proposed method is compared with support vector machine (SVM) algorithm. The result shows that the prediction accuracy of the proposed algorithm is much higher than SVM, which is of great importance in the field of traffic flow prediction.