Tensor based missing traffic data completion with spatial–temporal correlation

• We use tensor pattern to fully cover the spatial–temporal correlation of corridor freeway traffic volume.
• Various tensor patterns are tested on the traffic volume data collected from PeMS open database.
• We suggest that the tensor based method with spatial correlation achieves better performance than that without spatial information.
• The experimental results show that the proposed method can address the extreme case where the data of a long period of one or several weeks are completely missing.

Missing and suspicious traffic data is a major problem for intelligent transportation system, which adversely affects a diverse variety of transportation applications. Several missing traffic data imputation methods had been proposed in the last decade. It is still an open problem of how to make full use of spatial information from upstream/downstream detectors to improve imputing performance. In this paper, a tensor based method considering the full spatial–temporal information of traffic flow, is proposed to fuse the traffic flow data from multiple detecting locations. The traffic flow data is reconstructed in a 4-way tensor pattern, and the low-n-rank tensor completion algorithm is applied to impute missing data. This novel approach not only fully utilizes the spatial information from neighboring locations, but also can impute missing data in different locations under a unified framework. Experiments demonstrate that the proposed method achieves a better imputation performance than the method without spatial information. The experimental results show that the proposed method can address the extreme case where the data of a long period of one or several weeks are completely missing.