A0563
Title: Asymptotic properties of change point detection in high-dimensional data with a strongly spiked eigenvalue structure
Authors: Kento Egashira - Tokyo University of Science (Japan) [presenting]
Kazuyoshi Yata - University of Tsukuba (Japan)
Makoto Aoshima - University of Tsukuba (Japan)
Abstract: The aim is to consider detecting change points in high-dimensional data with limited sample sizes, particularly under a strongly spiked eigenvalue (SSE) model. A multivariate CUSUM-type statistic is introduced, designed to compare the means before and after each potential change point. Unlike many existing techniques, the approach avoids imposing sparsity constraints. The asymptotic behavior of the proposed statistic is investigated under the null hypothesis of no structural change, and the consistency of the corresponding change-point estimator is demonstrated under mild regularity conditions. In addition, the null distribution of the test statistic is derived specifically under the SSE setting. Extensive numerical experiments confirm the practical utility and robustness of the proposed methodology.
