Micro–Macro Changepoint Inference for Periodic Data Sequences

Existing changepoint approaches consider changepoints to occur linearly in time; one happens after another and they are not linked. However, data processes may have regularly occurring changepoints, for example, a yearly increase sales of ice-cream on the first hot weekend. Using linear here will miss more global features such as decrease due other product availability. Being able tease these from local (periodic) ones is beneficial inference. We propose periodic model this behavior using mixture time perspective. Built around Reversible Jump Markov chain Monte Carlo sampler, Bayesian framework used study behavior. To identify optimal positions we integrate into pruned exact (PELT) search algorithm. demonstrate that method detects both with high accuracy simulated motivating applications share Due micro–macro nature analysis, visualization results can be challenging. additionally provide unique perspective visualizations sequences. Supplementary Materials article available online.