Periodic data and changepoints: New methodology inspired by digital health applications
Conference
64th ISI World Statistics Congress - Ottawa, Canada
Format: CPS Abstract
Keywords: change-points, nonstationary time series, segmentation, timeseries
Session: CPS 13 - Statistics and health I
Monday 17 July 4 p.m. - 5:25 p.m. (Canada/Eastern)
Abstract
Traditional changepoint approaches consider changepoints to occur linearly in time; one changepoints happens after another and they are not linked. However, there exists fixed periods, e.g. days or years, which are repeated across time which results in periodic behaviours. Applying traditional changepoint approaches to these data sets will result in suboptimal solutions as they fail to utilise the periodic nature of these data processes. We propose a deterministic changepoint search method which utilises the periodic nature of these data processes by treating the time axis as circular. Furthermore, we extend this method to detect data processes which exhibit long term changes to periodic behaviour. We demonstrate these methods detect both periodic and global changepoints with high accuracy on motivating digital health applications.