ADPD: Anomaly Detection for Population Distribution in Geo-Space Using Mobile Networks Data

Real-time population mobility pattern at a specific time is an essential indication for sudden events. With the high penetration ratio of mobile phones, the mobile network could serve as a sensor network to monitor the population distribution, and user positioning without extra cost because the distribution of mobile phones is approximately the same as the population distribution. The abnormal spatial-temporal transition of the population is one of the critical indicators for sudden events. In this article, using the log data obtained from the mobile networks, we propose an AI-based framework, anomaly detection for population distribution (ADPD), to detect abnormal population distribution in geo-space. Different from the previous works, to detect anomaly of a specific grid area, the ADPD uses only the population information in the grid, which makes the ADPD more practical in the actual situation. We investigate the performance of the ADPD by running experiments based on the log data of the mobile network obtained during an actual sudden event, the 2018 Hualien Earthquake in Taiwan. Our study shows that the ADPD can identify the abnormal population transition grids nearby the grids with sudden events.