[3H5-GS-3] Data mining: Applied data mining (2)

[3H5-GS-3-02] Online anomaly detection of disasters using large-scale smartphone gps data

〇Ippei Takasawa¹, Yuki Morioka¹, Satoshi Goto², Yuya Suzuki², Yusuke Kato², Hiroshi Yadohisa¹ (1. Doshisha University, 2. Agoop Corp.)

Keywords:Spatio-temporal data, Mesh, People flow data

Recently, several natural disasters have occurred, such as the Great East Japan, Kumamoto, and Osaka earthquakes. Interruption of communication caused by natural disasters is a very serious problem, as it may cause secondary disasters, like local population concentration. Thus, when disasters occur, grasping overcrowding of population is an important task. Previous studies, estimated people flow using particle filter, and anomaly detection using state space model. However, they utilized only population data. Therefore, we conducted anomaly detection of mesh based on people flow data, considering both population and moving speed data. We calculated tendencies based on the data on the same hour at the same mesh and we identified the behavior of people flow in normal periods. Finally, we could calculate anomaly scores. The results indicate the usefulness of the proposed method applying people flow data of the past years events.