01659760

Component

This study proposes a large-scale Artificial Neural Network (ANN) model for predicting crashes on freeways and arterials in urban road networks. ANN is a biologically-inspired information processing paradigm which is composed of interconnected processing elements (neurons). This study considers the probability of crash occurrence as a class variable (output) and applies a ANN classifier to predict a crash occurrence within a given area in the network up to 3 hours into the future. As feature variables (input), this study uses traffic condition information (link speed) from the whole network. The proposed ANN model are trained and tested using actual traffic and crash data collected in Brisbane, Australia from 2013 to 2016. Crash prediction capabilities were compared with those from two other models: Logistic Regression and Support Vector Machine. The evaluation results show that the proposed ANN crash prediction model provides the best performance in all tested cases.

This paper was sponsored by TRB committee ABJ70 Standing Committee on Artificial Intelligence and Advanced Computing Applications.

18-06519

English

8p

2018

Transportation Research Board 97th Annual Meeting

Digital/other

Figures; References; Tables

Arterial highways; Crash risk forecasting; Freeways; Mathematical prediction; Neural networks; Probability; Regression analysis; Urban highways

Brisbane (Australia)

Highways; Planning and Forecasting; Safety and Human Factors

Transportation Research Board Annual Meeting 2018 Paper #18-06519

TRIS, TRB, ATRI

Jan 8 2018 11:41AM