Framework for Unveiling Change Points in Multivariate Signals with Non-Gaussian Patterns

This study focuses on the challenge of monitoring and managing complex systems by distinguishing sequential states while observing multiple variables. We introduce a methodology for change point detection in multivariate data with nonGaussian distribution based on fusion techniques, underpinned by multivariate statistical test based on the Cramer-von-Mises approach. To evaluate the performance of our method, we conducted a comparative analysis with established baseline techniques, namely e-Divisive and Kernel Change Point Analysis methods, using a multivariate sub-Gaussian distribution. Finally, we demonstrate the practical applicability of our approach by showing its ability to reduce invasiveness in detecting intracranial hypertension events during neurointensive monitoring of traumatic brain injury patient by identifying the temporal distribution structure in multivariate data.