A Change Dynamic Model for the Online Detection of Gradual Change

In the classic literature of change-detection, changes in the statistical properties of a stochastic process are assumed to occur via change-points, which demark instantaneous moments of complete and total process transition. In contrast many real world processes undergo such changes gradually. With this observation in mind, we introduce a novel change-dynamic model for the online detection of gradual change, in which classical change-points are identified in a hierarchal model. On both real and synthetic data we find that this model can allow for faster and more accurate identification of gradual change than traditional change-point models allow, and investigate empirically how delay in detection of this gradual change relates to alarm confidence.