Crash Prediction for Freeway Work Zones in Real Time: A Comparison Between Convolutional Neural Network and Binary Logistic Regression ModelAuthor/Presenter: Wang, Junhua; Song, Hao; Fu, Ting; Behan, Molly; Jie, Lei; He, Yingxian; Shangguan, Qiangqiang
Safety of drivers in freeway work zones has been a problem. Real-time crash prediction helps prevent crashes before they happen. This paper looks at real-time crash prediction in freeway work zones by using machine learning approaches. Both the Convolutional Neural Network and the Binary Logistic Regression model are introduced. For training and testing the models, crash data and traffic data from several freeways in D7 zone, Los Angeles, California, were used. Crash data were collected from California Highway Patrol Incident System, and traffic data were obtained from the Caltrans Performance Measurement System. Data processing and matching were conducted. Both the two models were trained and tested. Results show that the Convolutional Neural Network performed slightly better over the Binary Logistic Regression model in predicting crashes with a global accuracy of 79.50%. Despite this, the main merit of the Binary Logistic Regression model is that it is able estimate the impact of affecting variables on the probability of crashes and can help identify the factors related to risks in work zones. Machine learning approaches applied in this study perform well in crash prediction. In general, machine learning techniques and reliable real-time crash prediction applications can be promising in helping drivers and transportation engineers make timely responses to potential crashes on freeways.
Publication Date: June 19, 2021
Full Text URL: Link to URL
Publication Types: Reports, Papers, and Research Articles
Topics: Crash Data; Crash Risk Forecasting; Machine Learning; Mathematical Models; Risk Analysis; Traffic Data; Work Zone Safety; Work Zones