BACP: Bayesian Augmented CP Factorization for Traffic Data Imputation

Traffic data possesses spatiotemporal characteristics and encounters missing value problems due to sensor failure in real-world scenarios. Addressing this challenge requires a fast and efficient traffic data imputation method capable of leveraging spatiotemporal information. This paper proposes a Bayesian Augmented CP factorization (BACP) model for the traffic data imputation, which combines the Multiplicative Gamma Process (MGP) with the CP factorization to address the CP rank estimation. Extensive experiment results demonstrate that the BACP model has superior imputation accuracy. Additionally, it offers explicit interpretation of traffic patterns and exhibits lower computational complexity than other Bayesian methods.