Automatic incident detection algorithm based on under-sampling for imbalanced traffic data

Traffic data are highly skewed with rare traffic incidents in the real word, while most of the existing Automatic Incident Detection (AID) algorithms suffer from many limitations because of their inability to detect incidents under imbalanced traffic data set condition. Feasible AID algorithms based on under-sampling were proposed to process the imbalanced traffic data. An improved undersampling method based on the nearest-neighbor cleaning rule and Support Vector Machine (SVM) are combined to detect incidents. In terms of the optimization of SVM parameters, grid search method and Particle Swarm Optimization (PSO) algorithm were compared to obtain better detection performance. In addition, the effect of the number of nearest neighbors on detection performance was investigated. The I-880 data set was finally used in experiments to verify the proposed algorithms. The experimental results indicate that PSO algorithm is more competitive than grid search method for SVM parameter optimization. Moreover, the proposed AID algorithm based on under-sampling can achieve better performance.