Rule-Based Forecasting of Traffic Flow for Large-Scale Road Networks

Abstract

As traffic data collection becomes less costly and more commonplace, large-scale traffic flow forecasting is increasingly needed. This paper proposes a rule-based approach for forecasting traffic flow based on the K nearest neighbor (KNN) nonparametric regression model, the rule-based KNN (RKNN) model. Rules were extracted from the historical data through the use of rough set theory, which found the nearest neighbors. Traffic impact factors, such as weather and time of day, were incorporated into the rules. Every historical record was labeled with a rule. With current data on traffic flow states and traffic flow impact, the nearest neighbors could be found quickly from the historical data records covered by the corresponding rule. An additional methodology was proposed to keep the historical data and the rules up to date. A case study on an Interstate freeway in Virginia, I-395, was conducted to evaluate the performance of the RKNN approach. The results showed that the proposed approach could decrease the mean absolute percentage error by 26.86%. Moreover, the proposed algorithm reduced calculation time by 65.69%, compared with the traditional KNN algorithms. This difference indicates the effectiveness of the proposed algorithm for use with large urban road networks.