Prediction Time Horizon and Effectiveness of Real-time Data on Short-term Traffic Predictability

Although extensive work in short-term traffic prediction has been done, study on the predictability of short-term traffic which is the fundamental knowledge for selecting appropriate prediction model and evaluating its performance has received much less attention. Generally, predictability drops with the increase in prediction time horizon and the decrease in the effectiveness of real-time data. The decrease continues and converges to a deterministic mean that is generated by most historical methods. This study quantitatively examines the relationship among prediction time horizon, effectiveness of real-time data, and short-term traffic predictability based on traffic flow spatial-temporal relationship. Understanding traffic predictability can not only benefit the selection of prediction model, but also the identification of parameters of interest.