Abstract

Blind segmentation of time-series is necessary in many sensing applications where the system may switch between different states. In lifestyle sensor data there may be countless states such as sleeping, walking, working or exercising. The goal is to find the change-points in the time-series without additional information about the states. We compare several recent algorithms for change-point detection and propose a new hierarchical method that is computationally significantly more efficient than the best state-of-th-art method.