Modeling when and where a secondary accident occurs

Area of Science:

• Traffic Safety Engineering
• Transportation Systems Analysis
• Machine Learning Applications in Transportation

Background:

• Secondary accidents significantly contribute to traffic congestion and pose road safety risks.
• Effective traffic incident management necessitates robust strategies for secondary accident prevention.
• Understanding the spatiotemporal dynamics of secondary accidents is key to mitigation.

Purpose of the Study:

• To investigate the location and time of potential secondary accidents following an initial traffic incident.
• To develop and compare predictive models for the time and distance gaps between initial and secondary accidents.
• To identify the most effective machine learning approach for forecasting secondary accident timing.

Main Methods:

• Utilized a three-year dataset of accident and traffic loop data from California interstate freeways.
• Introduced a shock wave-based method to identify secondary accidents.
• Implemented and compared a linear regression model, a back-propagation neural network (BPNN), and a least squares support vector machine (LSSVM) using various influencing factors.

Main Results:

• The linear regression model showed poor predictive performance and goodness-of-fit.
• Both BPNN and LSSVM demonstrated adequate goodness-of-fit, with BPNN showing superior correlation (CORR) and lower mean squared error (MSE).
• The BPNN model excelled in predicting the time gap but struggled with distance prediction, outperforming the LSSVM in time forecasting.

Conclusions:

• The BPNN model is a viable tool for forecasting the time gap between initial and secondary accidents.
• Accurate time gap prediction can assist decision-makers and incident management agencies in preventing secondary collisions.
• Further research is needed to improve the prediction of the distance gap for comprehensive secondary accident mitigation.