Manifold Regularized Tucker Decomposition Approach for Spatiotemporal Traffic Data Imputation

Spatiotemporal traffic data imputation (STDI), estimating the missing value from partially observed traffic data, is an inevitable and challenging task in data-driven intelligent transportation systems (ITS). Due to the traffic data's multidimensionality, we transform the traffic matrix into the 3rd-order tensor and propose an innovative manifold regularized Tucker decomposition (ManiRTD) model for STDI. ManiRTD considers the sparsity of the Tucker core tensor to constrain the low rankness and employs manifold regularization and the Toeplitz matrix to enhance the model performance. We address the ManiRTD model through a block coordinate descent framework under alternating proximal gradient updating rules with convergence-guaranteed. Numerical experiments on real-world spatiotemporal traffic datasets (STDs) demonstrate that our proposed model is superior to the other baselines under various missing scenarios.