Bagging-SVMs Algorithm-Based Traffic Incident Detection

Real-time traffic incident detection is essential for traffic management and control. This paper proposes a traffic incident detection method based on a data mining technique named bagging support vector machine (bagging-SVMs). The performance of the proposed method was evaluated using the data collected on I-880. The performance measurements include detection rate (DR), false alarm rate (FAR), correct rate (CR), and F-measure. Several key issues were investigated to find the kernel function that better fits the new model and to calculate the important parameter bagging times. The results indicated that the standard SVM model with linear kernel function has advantages in traffic incident detection and is better than the other three models, comparatively. However, the bagging-SVMs model with a radial kernel function has a better comprehensive performance than the standard SVM model when bagging time is greater than 70. Models that use the polynomial kernel function also perform well even when the bagging time is less than 70. Preliminary results show superior performance of bagging-SVMs compared with legacy algorithms.