Potentialities of Data-Driven Nonparametric Regression in Urban Signalized Traffic Flow Forecasting

Single-interval forecasting of traffic variables plays a key role in modern intelligent transportation systems (ITSs). Despite the achievements of advanced ITS forecasting in literature, forecast modeling of urban signalized traffic flow, which shows rapid-intensive fluctuations associated with the nonlinear and nonstationary behavior of temporal evolution, is still one of its big challenges. From the perspective of field experts, the mathematical complexity of an advanced model is also a renewal obstacle in practice. On the other hand, the accessibility of large volumes of historical data and the concurrent advanced data management systems used to access them provide data-driven nonparametric regression with a renewal opportunity in practice. In order to address these problems effectively, this paper proposes a k nearest neighbor nonparametric regression (KNN-NPR) forecasting methodology to be tested against vast quantities of real traffic volume data collected from urban signalized arterials. The results show that the KNN-NPR model is clearly superior to two parametric models, Kalman filtering and seasonal autoregressive integrated moving average (ARIMA), in terms of both prediction accuracy and the construction of the directionality of temporal state evolution without a time-delayed response. Consequently, it appears that KNN-NPR, even though it is very simplified, is able to efficaciously capture the complex behavior of urban signalized traffic flow. (C) 2014 American Society of Civil Engineers.