An adaptive cluster-based sparse autoregressive model for large-scale multi-step traffic forecasting

Traffic forecasting has been extensively studied due to its importance for the design and development of Intelligent Transportation Systems (ITS). Most of the existing relevant literature focuses, almost exclusively, on the effectiveness of the traffic forecasting models, while neglecting the importance of computational efficiency. However, the need for faster models becomes increasingly urgent as the volume of available traffic data increases. In this paper, an effective and efficient model for large-scale multi-step traffic forecasting is presented. In particular, the classic autoregressive model is reformulated based on the idea that not all past traffic values are important for predicting future traffic values, and thus only some of them should be taken into account in the forecasting process. The selection of the appropriate past values is performed by the application of an eligibility criterion, controlled by a respective hyperparameter and its value is optimized using an efficient cluster-based method. The overall modelling approach leads to a sparse least squares problem, which is efficiently solved using a novel explicit preconditioned iterative method based on generic approximate sparse pseudoinverse. Large scale evaluation experiments were conducted using two real-world traffic datasets, and the experimental results indicate that the proposed model can achieve better balance between forecasting accuracy and computational efficiency compared to benchmark models.