Parallel SMO for Traffic Flow Forecasting

Abstract

Accurate traffic flow forecasting is crucial to the development of intelligent transportation systems and advanced traveler information systems. Since Support Vector Machine (SVM)have better generalization performance and can guarantee global minima for given training data, it is believed that SVR is an effective method in traffic flow forecasting. But with the sharp increment of traffic data, traditional serial SVM can not meet the real-time requirements of traffic flow forecasting. Parallel processing has been proved to be a good method to reduce training time. In this paper, we adopt a parallel sequential minimal optimization (Parallel SMO) method to train SVM in multiple processors. Our experimental and analytical results demonstrate this model can reduce training time, enhance speed-up ratio and efficiency and better satisfy the real-time demands of traffic flow forecasting.

Traffic Flow Forecasting Based on Combination of Multidimensional Scaling and SVM

“…Many scholars have been studying on the topic and many forecasting models have been developed. Commonly used methods include average method, ARMA, linear regression, nonparametric regression, and neural networks [1][2][3]. The forecasting precisions of these methods usually can't meet with the practical requirement.…”