A Machine Learning Method for Saturation Headway Estimation Based on Commonly Available Information

Estimation of saturation headway at signalized intersections is of vital importance for traffic supply measurement and signal timing optimization. Most previous studies either need manual participation in collecting label values or require dynamic or aggregated data (e.g., heavy vehicle and turning ratio) to build estimation model. In this paper, we proposed a machine learning method for saturation headway estimation using commonly available features. To efficiently extract reliable saturation headway for model training, an aggregate-based signal cycle identification method was developed to obtain saturation headway sequence automatically by the designed rule. The results show the extreme values at the interior and the end of the continuous headway sequence are identified accurately. The significance of selected features is proved by KS test. Meanwhile, the performance in diverse cases validates the effectiveness and robustness of the proposed model.