HRST-LR: A Hessian Regularization Spatio-Temporal Low Rank Algorithm for Traffic Data Imputation

• DOI: 10.1109/tits.2023.3279321
• Publication Date: 2023-06-08T17:43:46Z
• Authors: Xiuqin Xu; Mingwei Lin; Xin Luo; Zeshui Xu
• Keywords: Hessian matrix; Regularization

Intelligent Transportation Systems (ITSs) are vital for alleviating traffic congestion and improving efficiency. Due to the delay of network transmission failure detectors, massive missing data often produced in ITSs, which evidently decreases accuracy decision-making road management. Hence, how establishing a precise efficient estimation becomes hot yet thorny issue. Low-rank matrix completion (LR-MC) model has proven be highly effective address this issue owing its fine representativeness such high-dimensional incomplete data. However, existing LR-MC models mostly fail inherently temporal spatial correlations hidden structure, resulting low accuracy. To improve it, paper proposes Hessian regularization spatio-temporal rank (HRST-LR) algorithm with three main-fold ideas: a) imposing low-rank property into global features precisely learning b) capturing evolvement via second-order difference time-series constraint, c) modeling similar space segments through thus exploring local correlation between representing patterns Experimental results on four sets prove that HRST-LR outperforms several state-of-the-art methods root mean squared error improvements higher than 14% when rate is 90%. valuable imputation need performing analysis.