Decision tree learning for freeway automatic incident detection

In this research, the technique of decision tree learning was applied to cope with traffic incident detection problem. The traffic data containing volume, speed, time headway and occupancy at both upstream and downstream detectors for testing were generated with a traffic simulation system. The performance of automatic incident detection (AID) models is evaluated based on detection rate, false alarm rate, mean time to detection, classification rate, as well as the receive operating characteristic curves. The detection performance of the decision tree was compared to neural networks which yield superior incident detection performance in the previous studies. The experimental results indicate that decision tree is competitive with neural networks, and the operation of discretizing attribute can enhance detection rate. Besides, derived data was employed to deal with the influence of road geometric characteristic. The conducted experiment indicates that these two operations is helpful for AID and can improve the performance of detection.