Adopting Machine Learning Methods to Predict Red-light Running Violations

Statistics demonstrate that a large number of crashes occur at signalized intersections due to traffic violations, specifically red light running (RLR). In order to prevent/mitigate intersection-related crashes, these violations need to be identified before they occur, so appropriate actions can be taken. Several factors such as vehicle speed, Time to Intersection (TTI), Distance to Intersection (DTI), age, gender, etc. influence the drivers behavior when approaching intersections. However, the driver-related factors (i.e. age, gender) are more difficult to obtain in practice. On the other hand, kinetic factors (e.g. speed, acceleration) can be obtained by monitoring the movement of vehicles through video cameras installed on the infrastructure or through on-board devices installed on the vehicles. Hence, the problem of interest is to develop models to predict RLR violations using kinetic information of vehicles. A monitoring period was defined to extract data from each vehicle before reaching the intersection. Machine learning techniques, namely Support Vector Machine (SVM) and Random Forest (RF), were adopted to develop prediction models. The minimum Redundancy Maximum Relevance (mRMR) feature selection method was used to identify the most important factors for model development. To evaluate the performance of the models the K-fold cross-validation and out-of-bag (OOB) errors were used for the SVM and RF models, which contributed to high prediction accuracies of 97.9 and 93.6 percent, respectively. It was shown that other than the critical instant at which the traffic signal changes to yellow, an appropriate monitoring period with respect to the yellow onset can provide additional useful information ensuring that the driver decision occurs during that period.