A online boosting approach for traffic flow forecasting under abnormal conditions

In this paper, we propose a online boosting non-parametric regression (OBNR) model for traffic flow forecasting, which can work effectively under abnormal traffic conditions. The model is composed of two part: the base part and the boosting part. The base part deals with normal prediction, while the boosting part constructed in a gradient boosting way adapts the model with abnormal conditions and updates in real time. When the traffic state turns back to normal, the boosting part is disabled and the base part works well again. Experiments on highway station output flow show that OBNR is much more effective than traditional online learning models in dealing with abnormal traffic conditions.