A STAR model for urban short-term traffic flow forecasting

In this paper a nonlinear univariate time series model, the smooth transition autoregressive (STAR) model, is proposed to forecast short-term traffic flow. The empirical study is carried out to investigate multi-step-ahead out-of-sample forecast performance of the nonlinear univariate model and a linear one, namely an ARIMA model used as a benchmark model for comparison. The results indicate that the STAR model can better match the traffic behaviour of extreme peaks and rapid fluctuation, and thus it performs better especially when there are sudden shifts in short-term traffic flow volatility.