Developing a new real-time traffic safety management framework for urban expressways utilizing reinforcement learning...

Area of Science:

• Traffic Engineering
• Artificial Intelligence
• Data Science

Background:

• Urban expressways face significant crash frequencies and losses.
• Real-time crash risk prediction (RTCRP) and automatic incident detection (AID) are crucial for proactive traffic management.
• Existing models struggle with large, disaggregated datasets from new transport technologies.

Purpose of the Study:

• To propose a novel reinforcement learning tree (RLT) approach for RTCRP and automatic crash detection (ACD).
• To develop a real-time traffic safety management framework for urban expressways using online traffic data.
• To evaluate the RLT model's performance against traditional machine learning techniques.

Main Methods:

• Developed RTCRP and ACD models using a state-of-the-art reinforcement learning tree (RLT) approach.
• Integrated recorded traffic flow and historical crash data for model development.
• Compared RLT performance with logistic regression (LR), support vector machine (SVM), and deep neural network (DNN).

Main Results:

• The RLT approach significantly outperformed LR, SVM, and DNN in developing RTCRP and ACD models.
• The proposed framework achieved approximately 96% correct crash prediction/detection with a 10% false-alarm rate.
• Model performance improves with increased data and additional predictive factors; a minimum of 20 observations per variable is recommended for RLT training.

Conclusions:

• The RLT approach offers a superior method for real-time traffic crash prediction and detection on urban expressways.
• The developed framework enhances traffic safety management through accurate and timely crash identification.
• Data quantity and factor inclusion are critical for optimizing predictive performance in traffic safety models.