Analysis and prediction of intersection traffic violations using automated enforcement system data.

Area of Science:

• Traffic safety research
• Data analytics in transportation
• Predictive modeling for road safety

Background:

• Automated enforcement systems (AES) supplement traditional traffic enforcement.
• AES data offers valuable insights for traffic safety research and analysis.
• Understanding factors influencing traffic violations is crucial for improving road safety.

Purpose of the Study:

• To analyze factors influencing traffic violations using data from AES.
• To predict the probability of traffic violations at intersections.
• To evaluate the performance of machine learning models in traffic violation prediction.

Main Methods:

• Utilized 24 independent factors encompassing time, space, traffic, and weather.
• Employed logistic regression for initial factor analysis.
• Applied random forest algorithm for violation probability prediction, comparing it with other models.
• Implemented proximity weighted synthetic oversampling (ProWSyn) to address data imbalance.

Main Results:

• Midday periods, weekends, residential areas, collector roads, traffic congestion, high traffic flow, low wind speed, and low temperatures increase violation probability.
• Random forest algorithm outperformed logistic regression, Gaussian naive Bayes, and support vector machine in prediction.
• ProWSyn technique improved model performance, with the random forest achieving an AUC of 0.914 and OOB error of 0.0787.

Conclusions:

• Traffic violation data from AES can be effectively used for safety research.
• The random forest algorithm, enhanced with ProWSyn, demonstrates superior performance in predicting traffic violations from imbalanced datasets.
• Identifying key influencing factors aids in developing targeted safety interventions.