B0523
Title: Categorical changepoint detection for activity sequences
Authors: Jessica Gillam - Lancaster University (United Kingdom) [presenting]
Rebecca Killick - Lancaster University (United Kingdom)
Simon Taylor - Lancaster University (United Kingdom)
Jamie-Leigh Chapman - Howz (United Kingdom)
Abstract: Age UK state, as of April 2019 there are close to 12 million people in the UK aged 65 or over, of which 3.8 million live by themselves. Around 40\% have a long term condition and approximately 30\% need help with at least one daily activity. There is an increasing body of research that indicates changes in daily routine signal a change in health and well-being. This project is in collaboration with Howz who are using sensors on common household appliances to automatically detect changes in well-being. The goal is to investigate which appliances are being used and in what order to detect a change in behaviour whilst allowing for an individual's daily variation. For example, a move from using the hob to microwave might indicate a change in confidence levels in preparing meals. Classic pattern recognition literature finds it difficult to classify changes in noisy categorical data, whilst the majority of the changepoint literature focus on numerical time series. We present a new method to detecting changes in patterns of behaviour using activity event based sequences with the aim to detect changes on a day-to-day basis.
