Traffic pattern forecasting using time series analysis between spatially adjacent sensor clusters

In most US cities, the traffic monitoring networks are used to sense the real-time traffic. Such information helps drivers to select routes and assists traffic control agencies. In this paper, we propose a new approach that extends such systems by forecasting future traffic using the real-time sensor inputs. Our approach has two features. First, it predicts the shape of the future traffic episodes along with their values. Second, our approach explores the temporal relationship between adjacent sensor groups. The predictions are achieved between two adjacent sensor groups and are used as evidences to achieve further predictions on non-adjacent sensor groups. Our experimental results show that our approach achieves an average prediction accuracy up to 80%, whereas the extension of existing linear regression based method only achieve an average accuracy of 36%.