Machine learning methods to quantify the physical behaviour of older adults and a study of healthy ageing

Abstract

This thesis investigates how wearable sensors and machine learning can be used to better measure physical behaviour in older adults and to understand how this relates to healthy ageing. Using data from two studies, the research developed and evaluated machine methods for recognising daily activities and estimating energy expenditure from sensor signals. Deep learning models, especially recurrent neural networks, improved performance over more traditional approaches and enabled more accurate and scalable monitoring of behaviour. The thesis also examined how changes in physical activity relate to health outcomes during a lifestyle intervention. The findings show that even modest increases in daily activity are associated with beneficial changes in health markers, while also revealing important differences between men and women. Overall, this work shows that combining wearable technology with advanced data-driven methods can improve behavioural measurement and provide useful insights for promoting healthy ageing.