A Noise-eliminated Gradient Boosting Model for Short-term Traffic Flow Forecasting

Abstract

Accurate and real-time short-term traffic flow forecasting is an important prerequisite for traffic guidance and control. A single forecasting method is difficult to handle all the periodicity, nonlinearity, and randomness of traffic flow. The traditional machine learning algorithm is difficult to achieve parallel computing. In this paper, a traffic flow forecasting method based on wavelet denoising and XgBoost hybrid model is proposed. The discrete wavelet denoising algorithm is first employed to preprocess the traffic flow data. Then, the XGBoost algorithm is fed by the denoised historical traffic flow data to predict traffic flow. It not only keeps the trend of traffic flow in each sampling period but also reduces the impact of noise. The proposed traffic flow forecasting method is evaluated on four benchmark dataset and compared with the state-of-the-art models. The results show that the prediction accuracy of the algorithm is much higher than that of control models.