THE BEHAVIOR CHANGE IN INTER-PREFECTURE TRAVEL UNDER THE COVID-19 PANDEMIC: DETECTION BY SPARSE NON-NEGATIVE MATRIX FACTORIZATION

Abstract

 This study reveals the changes in inter-prefecture travel behavior under COVID-19 pandemic. The mobile phone location data is useful for this analysis because it enables us to obtain accurate and temporally detailed inter-prefecture travel data. We applied non-negative matrix factorization to this high-dimensional data for understanding the time-series features. Furthermore, by adding a penalty term of L₁ norm to the model, it is possible to extract more distinct features. As a result of this analysis, we have quantitatively identified the following behavioral changes. This analysis revealed the following quantitative behavioral changes: inter-prefecture commuting behavior in the Tokyo metropolitan area, weekend travel, and returning home during long vacations. Furthermore, the residual information revealed that these behavioral changes were spatially unevenly distributed.