Improving the Traffic Data Imputation Accuracy Using Temporal and Spatial Information

The missing data problem exists in a lot of transportation systems while many traffic applications require accurate traffic data to work well. So, various methods were proposed for traffic data imputation in recent years. However, most of these approaches only consider the information of a single detector. In this paper, we use the temporal and spatial information collected from multiple detectors to improve the traffic data imputation accuracy. Specially, we consider Probabilistic Principal Component Analysis (PPCA) and Mixed Probabilistic Principal Component Analysis (MPPCA), which had been proven to be useful methods to handle data imputation. We examine the performance of these two methods using the temporal and spatial information. Tests show that when only using the spatial information, the errors of these two methods are reduced and MPPCA works even better than PPCA. Moreover, when both of the temporal and spatial information from multiple detectors are used, more imputation accuracy is obtained than only using the spatial information for MPPCA.