Spatial-temporal traffic speed bands data analysis and prediction

The development of Intelligent Transportation System (ITS) requires insightful information and predictions from spatial-temporal traffic data. In order to improve commuters' experiences by forecasting and proactively taking actions on city traffic, in this paper, we analysed temporal and spatial patterns of low-resolution traffic speed bands and predicted future traffic status based on SVM learning models. Real speed bands data of Singapore were used for analysis and prediction. The performance of machine learning models were evaluated by speed bands data of road segments in Central Business Districts (CBD) and Tampines area at Singapore. The results demonstrated that multiple kernel learning (MKL) combining both temporal and spatial patterns showed better performance in predicting next 5 minutes and 1 hour traffic conditions in both CBD and rural areas (Tampines). In short-term prediction (next 5 minutes), SVM learning using temporal pattern showed better performance in both CBD and Tampines than using spatial patterns. However, the long-term prediction (next 1 hour) of CBD area has lower accuracy than rural area (Tampines). In general, we can achieve up to 88% prediction accuracy of traffic speed band at entire Singapore.