Road Traffic State Prediction with a Maximum Entropy Method

Abstract

The prediction of the traffic state can give the people the important traveling information. In this paper, the traffic state prediction problem is studied. A Maximum Entropy(ME) approach is proposed for the traffic state prediction, which consider the prediction process as a classification problem instead of predicting the traffic flow parameters. The traffic state is defined as six classes according to the level of service. The Maximum Entropy approach is introduced to model this prediction process. In the ME framework, more different features can be used regardless of the features' dependence. The temporal and spatial features can be used together, which is hard to complished in the previous methods. The experiments show that the Maximum Entropy model is competent for the traffic state prediction. The most advantage of the Maximum Entropy model is that the road network features can be introduced. And this method can be also introduced to predict the long time traffic state in the future work.