Multi-source Traffic Data Calibration with Optimized Adaboost

Abstract

Abstract.A large amount of real-time traffic data supports the processing requirements of traffic state discrimination and prediction. Therefore, accurate real-time traffic information can be grasped for effective detection of outliers. In this paper, an optimized AdaBoost model for screening abnormal traffic samples is proposed based on the multi-source features of the detected data. Considering the unbalanced characteristics of traffic data, AdaBoost is optimized by cost-sensitive method, which avoids the problem that classification performance is degraded by non-equilibrium detection data. The accuracy, false alarm rate and false alarm rate of the model test are verified by the example of expressway test data set. The experimental results show that the AdaBoost model is 5.547% higher than the AdaBoost method in screening traffic samples. The algorithm can effectively adjust the classification error caused by unbalanced data.