Nonnegative Coupled Matrix Tensor Factorization for Smart City Spatiotemporal Pattern Mining

With the advancements in smartphones and inbuilt sensors, the day-to-day spatiotemporal activities of people can be recorded. With this available information, the automated extraction of spatiotemporal patterns is crucial to understand the people’s mobility. These patterns can assist in improving the smart city environments like traffic control, urban planning, and transportation facilities. The smartphone generated spatiotemporal data is enriched with multiple contexts and efficiently utilizing them in a Machine Learning process is still a challenging task. In this paper, we propose a Nonnegative Coupled Matrix Tensor Factorization (CMTF) model to integrate and analyze additional contexts with spatiotemporal data to generate meaningful patterns. We also propose an efficient factorization algorithm based on variable selection to solve the Nonnegative CMTF model that yields accurate spatiotemporal patterns. Our empirical analysis highlights the efficiency of the proposed CMTF model in terms of accuracy and factor goodness.