SASTC: Spatial-Aware Sparse Tensor Completion for Large-Scale Traffic Data Recovery

In Intelligent Transportation Systems (ITS), missing and deviated traffic data are ubiquitous. However, most existing studies fail to address both the imputation of missing data and the correction of deviated data simultaneously. Moreover, while much of the research focuses on traffic data recovery for localized areas, there is limited work on large-scale traffic data recovery. In this paper, we propose a novel approach called Spatial-Aware Sparse Tensor Completion (SASTC) for large-scale traffic data recovery. First, we construct a tensor model for SASTC, treating traffic data as a tensor and redefining the traffic data recovery problem as a tensor completion problem. Next, we present the SASTC model, which incorporates the Laplacian matrix and context matrix to capture spatial information and optimize the objective function, thereby enhancing model accuracy. Finally, we design a parallel algorithm for SASTC on GPU based on the Stochastic Gradient Descent (SGD) method to accelerate large-scale tensor completion computation. Extensive experiments demonstrate that the proposed method outperforms state-of-the-art techniques in various scenarios.