Skip to Main Content
Short-term prediction is one of the essential elements of intelligent transportation systems (ITS). Although fine prediction methodologies have been reported, most prediction methods with current time-series data lead to inefficient predictions when current or future time-series data either exhibit fluctuations or abruptly change. In order to deal with this problem, a dynamic multi-interval traffic volume prediction model, based on the k-nearest neighbour non-parametric regression (KNN-NPR), is introduced in this study. In an empirical study with real-world data, the input parameters of the proposed model including the k-values for the nearest neighbours in the neighbourhood and the dm-values for the number of lags were optimised according to the multi-interval prediction horizon in order to immediately capture the directionality of the future states and to minimise the prediction errors. The presented model performed effectively in terms of prediction accuracy, despite multi-interval schemes, to the same degree as applications of the real ITS, even if the time-series data abruptly varied or exhibited wide fluctuations. It can clearly be seen that the proposed methodology is one of the promising system-oriented approaches in the area of multi-interval traffic flow forecasting.