A0496
Title: Changepoint detection in periodic behaviour
Authors: Owen Li - Lancaster University (United Kingdom) [presenting]
Rebecca Killick - Lancaster University (United Kingdom)
Abstract: Traditional changepoint approaches consider changepoints to occur linearly in time; one changepoint happens after another, and they are not linked. However, data processes may exhibit periodic behaviour and so changepoints will occur regularly, e.g. sleeping patterns and daily routine behaviour. Using linear changepoint approaches in these settings will miss global changepoint features which affect changepoints on the more local (periodic) level, for example, the introduction of local lock-downs affecting sleeping patterns. Being able to tease these global changepoint features from the more local (periodic) ones is beneficial for inference. We propose a deterministic periodic changepoint method using a periodic (circular) time perspective. This is done by adapting the Segment Neighbourhood changepoint method to the periodic time perspective. We then integrate this local changepoint model into the pruned exact linear time (PELT) search algorithm to identify the optimal global changepoint positions. We demonstrate that the method detects both local and global changepoints with high accuracy on simulations and motivating digital health applications.
