A1253
Title: Point process model for human activity data: Detecting changepoints
Authors: Israel Martinez-Hernandez - Lancaster University (United Kingdom) [presenting]
Abstract: Monitoring changes in people's behaviour or activities is a vital tool for early health intervention, as these changes often signal underlying medical issues. However, natural circadian rhythms such as activity spikes during meals and rest at night create "noise" that can mask significant long-term trends. The proposed approach focuses on filtering out these expected within-day fluctuations to detect meaningful shifts in behavioural patterns from one day to the next. To achieve this, a full day of events is treated as a single observation, modelled as the realisation of a point process, similar to functional data analysis. By analysing the sequence of these point processes rather than individual event times, subtle changes in intensity that indicate potential health decline can be identified. This approach is particularly effective for home activity data, providing a framework that is both highly sensitive and easy to visualise in terms of trends and changes.
