Title: Change point detection in high-dimensional time series with both spatial and temporal dependence
Authors: Jun Li - Kent State University (United States) [presenting]
Abstract: High-dimensional time series are characterized by a large number of measurements and complex dependence, and often involve abrupt changes at unknown time points. We will present a new procedure to detect the change points from high-dimensional time series data. An asymptotic testing procedure is established for the hypothesis of existing any change point. When the null hypothesis is rejected, a binary segmentation method is conducted to estimate multiple change points. We will demonstrate the impact of sample size, dimensionality and the location of the change point on the proposed method. Compared with other methods, the proposed procedure allows both sample size and dimensionality to diverge without constraint on the growth rate of dimensionality. Moreover, it does not assume Gaussianity, and incorporates both spatial and temporal dependence without imposing restrictive structural assumptions.