Understanding human activity patterns in cities

Civil activities in a city are not random because of regular and routine behaviors of people. Recently, sensor networks and mobile phone networks enable us to collect those activities as spatio-temporal big data. However, because such data exhibit complex phenomena and are often incomplete due to sensor corruptions or network troubles, it is hard to understand civil activities by just visualizing spatio-temporal flow in a city from obtained data. We propose a machine learning method that discovers latent patterns of spatial activity patterns and its corresponding temporal dynamics. Furthermore, our method can recover missing entries in data with extracted latent patterns. We demonstrate that our method can analyze automobile traffic data in a city and extract more interpretable patterns than existing pattern extraction methods.

Please click the thumbnail image to open the full-size PDF file.