Weighed support vector machine for traffic incident detection

A weighed support vector machine(SVM), based on the importance of the samples, was proposed to solve the problem of low detection caused by imablanced samples. The weight of the samples was determined by their discriminate errors. The measured data was used to test the performance of the proposed algorithm. The basic SVM, the weighed SVM based on the sample number and the weighed SVM based on sample importance were tested under different imbalanced samples. The results show that the algorithm can determine the weight of the sample according the sample, which can improve the robust of the algorithm; the bigger the imbalance of the samples, the lower the detection ratio of all three algorithms. With the same samples, the ratio of detection of the basic SVM is the lowest and the weighed SVM based on sample importance is the highest; in traffic incident detecting, in order to improve detecting rates, the weighted SVM based on sample important is the best choice; under different unbalanced sample rates, detection effects are different; the higher the unbalanced rate, the worse the detection effect.