Abstract:
To address the issue of quantifying the predictability of traffic status, we proposed an entropy-based approach. First, the regularity of a traffic status sequence is evaluated in terms of entropy and converted to the corresponding predictability with a binary entropy function. Then, considering the dynamics of predictability, we introduced instantaneous entropy for quantifying the predictability of traffic status at specific time slot. In addition, we analyzed the correlation between the prediction performance of four representative traffic status prediction models and the quantified predictability of traffic status. As suggested by the experiments, the proposed approach can effectively quantify the predictability of traffic status from both static and dynamic views, and uncovered that different categories of prediction models behave differently against the predictability, hence providing meaningful supports for the selection and design of traffic status prediction models,