A1656
Title: On influential variables driving change points in high dimensional data
Authors: Shrog Albalawi - Durham university (United Kingdom) [presenting]
Reza Drikvandi - Durham University (United Kingdom)
Abstract: Detection of change points in a sequence of high dimensional observations is a challenging problem, especially when the change is due to a small number of variables, often known as a sparse change point. A question of interest is how to identify the variable or group of variables that caused a sparse change point in high-dimensional datasets. When a change point is detected, we propose a method that identifies the crucial variables driving the change point by grouping variables according to an appropriate distance measure and assessing how those groups contribute to the change point. The approach reduces dimensionality while accurately identifying the most influential variables. By applying a grouping strategy where the number of groups k satisfies $k < n < p$ with n being the number of observations and p the number of variables, the method enhances the interpretability of high dimensional data and provides insights into the factors driving the change points. Through numerical simulations, the performance of the method is illustrated. The results show an improved performance as both p and n increase, even in sparse settings.
