Abstract: In order to reduce traffic delays caused by traffic incidents, an new AID (automatic incident detection) algorithm, SVM-AID algorithm, was proposed based on support vector machines (SVM). Using actual traffic data of the I-880 database, the classification performance of the SVM-AID algorithm was tested, and the effects of the parameters in SVM on the classification were analyzed. The experimental results show that the parameters should be chosen carefully because they have great effects on the classification. The correct classification rate of the proposed algorithm is more than 98% and its mean time to detect is less than 5 s to indicate a better performance over other AID algorithms based on artificial neural networks.