[2P5-GS-10] AI application: detection / identification

[2P5-GS-10-05] A study of human mobility data structure for COVID-19 cases prediction

〇Yutaro Mishima^1,2, Guillaume Habault¹, Shinya Wada^1,2 (1. KDDI Research, Inc., 2. KDDI Corporation)

[[Online]]

Keywords:Spatio-temporal prediction, COVID-19, Location data, Human mobility data

In order to prevent the spread of COVID-19, many countries have taken measures to restrict people's movement, which led to a drastic reduction in economic activities. Thus, it is of crucial importance to maintain economic activities while preventing more infections. Recently, studies were conducted to predict the number of COVID-19 cases using data processed from people's location information. Japanese Mobile Network Operators also provide such human mobility data. However, usual anonymization processes (used to produce such data) remove meaningful information, such as transition information. Therefore, the question regarding the most effective human flow representation remains. In this paper, we conduct an experiment to predict the number of infected people in Tokyo through the year 2020 and 2021 using two types of human flow data. A usual grid-level human mobility data without transition information, and one focusing on inter-cell transition. Results show that, when transition information is included, prediction performance improves for all major waves of infection that year. With this experiment, we confirm the effectiveness of transition information when modeling spread of infection.