B1792
Title: Unlocking the potential of wearable data: Time series analysis for comprehensive understanding of physical activity
Authors: Melina Del Angel - University of Bath (Mexico) [presenting]
Matthew Nunes - University of Bath (United Kingdom)
Dylan Thompson - University of Bath (United Kingdom)
Abstract: Physical activity is important for the treatment and management of multiple health conditions. Understanding its relationship with certain diseases is key for health professionals to make tailored advice to patients, targeting specific health dimensions. Recently, wearable monitors have been introduced as a new technology for monitoring physical activity in a more reliable way than the usual self-reported methods. However, analyzing wearable data represents a new challenge because of its large volume, and non-stationarity. Thus, there is a need to develop proper methods for analysing wearable data. In the current literature, most of the methods used draw conclusions from mean-based methods, ignoring the time-dependent structure and overlooking important information regarding individuals' behaviour. One promising but unexplored approach for physical activity data is time series analysis. It is proposed to use the locally stationary Wavelet process + trend, a type of time series that can handle first and second-order non-stationarities and provides a curve trend estimation that can be used to assess individuals' performance. This model provides a more robust way to understand the underlying components of physical activity data and provides health professionals with new tools to analyze longitudinal medical data, complementing and enhancing conclusions drawn from conventional methods.