Abstract: A feature weighed support vector machine was proposed to solve for low detection of automatic incident detection caused by redundant feacture.Because the factors related to incident detection include occupy, traffic volume and average speed of the upstream and downstream, and the influence of each factor is different, the influence value was determined by the classe margin of the each feature.Using actual traffic data, the detection performance of the feature weighed SVM algorithm was tested.The results show that the weighed value of occupy is the biggest.It indicate that the influence of occupy is the biggest, which is consistent with the fact.For the same sample, the performance of the proposed algorithm is superior to the standard support vector machine (SVM) .