Utilizing Reinforcement Learning Tree to Develop the Real-time Traffic Safety Management Framework on Urban Expressways

How to reduce the crash frequency and the loss caused by crashes on urban 1 expressways is the main objective for traffic managers and researchers. Real-time crash risk 2 prediction (RTCRP) is one of the most important techniques to identify crash precursors so as to 3 take measures to smooth the traffic fluctuations; and automatic incident detection (AID) is 4 another important technique to identify the occurrence of incidents timely so as to take measures 5 to reduce the its negative impacts on traffic flow. Exploring better modelling methods is still the 6 important research point in this field. In this paper, a state-of-the-art reinforcement learning tree 7 (RLT) approach is proposed to develop RTCRP and AID models, and is further implemented to 8 build a traffic safety management framework on urban expressways with real-time traffic data 9 streaming. Historical crash data and corresponding traffic flow data were integrated and divided 10 into a training dataset and a test dataset to develop and test RTCRP models and AID models. In 11 addition, the prediction results were compared with those given by other frequently used 12 classification algorithms, including random forest and support vector machine (SVM). The 13 results prove that RLT slightly outperforms random forests and RLT can improve 3.6% and 1.8% 14 compared with the SVM in RTCRP and AID. At the cost of 10.0% false-alarm rates, 79.8% and 15 92.9% of crash cases can be identified and detected correctly by the RLT model. RLT has the 16 potential to predict and detect the crash occurrence in the traffic safety management. 17 18     «

How to reduce the crash frequency and the loss caused by crashes on urban 1 expressways is the main objective for traffic managers and researchers. Real-time crash risk 2 prediction (RTCRP) is one of the most important techniques to identify crash precursors so as to 3 take measures to smooth the traffic fluctuations; and automatic incident detection (AID) is 4 another important technique to identify the occurrence of incidents timely so as to take measures 5 to reduce the its negative impacts...     »