A traffic data imputation method considering multi-time characteristics

In order to solve the problem of missing traffic data due to various reasons, we propose a multiple temporal tensor factorization (MTTF) model based on the Bayesian pattern to impute missing data. This model transfers traffic data into a tensor form, taking into account the multiple time characteristics of the traffic data. The factorization tensor not only contains multiple time factors but also considers time and space characteristics of traffic data at the same time. Most importantly, it is closer to the real traffic situation. We conduct experiments on various situations of missing data, and the rate of missing data ranges from 10% to 90%. The parking occupancy data of Birmingham between 8:00 to 16:30 was used in the experiment from October 2020 to December 2020. Based on actual data verification, the result indicated that, the imputation effect of this model can perform better in the case of various missing rates compared to the same type of model.