Real-Time Traffic Incident Detection with Classification Methods

摘要

It is well known that traffic incident detection is essential to intelligent transportation system (ITS) and modern traffic management. Compared to traditional models based on traffic theory, some data mining computational algorithms are believed more appropriate and flexibility for automatic incident detection. In this paper, four classification models were introduced and their parameters were selected by tenfold cross-validation. Using an open dataset their predictive performance was compared based on five criteria. The results show that the classification models perform well to detect traffic incidents and no over-fitting problem. What's more, AdaBoost-Cart and Naive Bayes models seem to outperform support vector machine and Cart models since they provide superior detection rate. However, they cost long time to train.