CFE 2019: Start Registration
View Submission - CMStatistics
Title: Structural change detection via common principal component analysis Authors:  Tatsuya Matsukawa - Chuo University (Japan) [presenting]
Toshihiro Misumi - Yokohama City University (Japan)
Yoshihiko Maesono - Chuo University (Japan)
Sadanori Konishi - Kyushu University (Japan)
Abstract: The problem of detecting a covariance structure change for multivariate data collected over time is considered. In order to detect the structure change point, we propose a new approach based on common principal component analysis (CPCA). The CPCA enables us to visually capture a pattern extraction and structural change as time passes on common principal component axes by simultaneously diagonalizing the variance-covariance matrices of all groups. To objectively identify the covariance structure change, we introduce a new procedure based on a deviance for the CPCA model. Using the difference of the deviances between the full model and submodel under each common principal component assumption, we detect the change point of the covariance structure. We can also use the new method to reduce the dimension of the data. We investigate the effectiveness of our proposed method through the analysis of real data and a simulation study.