Support vector machine models for freeway incident detection

This paper presents the development of freeway incident detection models based on the recently developed support vector machine (SVM) classifier. The overall framework, algorithm development, implementation and evaluation of this technique are discussed. Freeway traffic flow parameters measured by sensors, such as occupancy and volume are used by the SVM models to detect incidents. The performance of the developed algorithms is evaluated using the common criteria of detection rate (DR), false alarm rate (FAR), mean time to detection (MTTD), and misclassification rate (MCR). A performance index (PI) is then calculated by combining these performance criteria. Offline test results using real data collected at the I-880 Freeway in San Francisco Bay area. California have shown that the SVM models produce better PIs compared to the multi-layer neural network models.