Analyzing COVID-19 Spread Mechanisms in Japan Using Time Series Decomposition, Clustering, and Regression

In this study, we analyzed daily time series data on newly confirmed COVID-19 cases in each prefecture of Japan to investigate the mechanisms driving the virus’s spread. The dataset spans from 20 January 2020 to 7 May 2023, covering 1204 days and providing insights into daily case variations across prefectures. First, the time series data for each prefecture were decomposed into trend, weekly variation, and short-term components. Using the trend components, we estimated the time lag of infection spread between prefectures, revealing that Okinawa and Tokyo consistently led the spread compared to other regions. Factors influencing these lag values were also analyzed. Through a cluster analysis, we categorized all of the prefectures into 13 groups and conducted a detailed investigation of the infection dynamics within each group. The results highlighted that regions centered around Tokyo in the Kanto area acted as a primary epicenter, driving the nationwide spread through regions centered around Osaka and Kyoto. Additionally, we examined the effects of holidays and seasonal variations within the short-term components using a regression analysis. The findings showed that holidays initially had a negative effect on case numbers, followed by a significant positive effect one week later. Regarding seasonal effects, November exhibited the highest positive impact, while March demonstrated a negative impact during the analyzed period.