A Hybrid Approach to Traffic Incident Management: Machine Learning-Based Prediction and Patrol Optimization

Abstract

The objective of Traffic Incident Management (TIM) is to respond to traffic accidents and other traffic incidents as immediately as possible, and to eliminate the consequences of such incidents with the utmost expediency. One of the most critical issues in traffic incident management is the problem of assigning police patrols. Despite the existence of numerous defined patrol areas within urban contexts, police teams are typically deployed to locations where accidents and traffic incidents are most likely to occur. The deployment of police patrols to the closest locations to potential accident sites in a dynamic traffic system is an effective strategy for reducing response times, minimizing carbon emissions, and lowering costs. This study presents a novel five-stage hybrid approach to the patrol assignment problem, which estimates the traffic accident-incident location, event time, and traffic incident duration based on machine learning algorithms. The successful prediction results are then used as inputs in the patrol assignment optimization model. The presented model was tested in a pilot region defined in Istanbul, and the results demonstrated that the machine learning-based model reduced the response time to traffic incidents, transportation costs, and was effective in reducing carbon emissions in the police assignment problem.