Measuring physical activity in obese populations using accelerometry

The thesis is concerned with objectively measuring human physical activity through \ud accelerometry, and compares the effectiveness of algorithms between obese and non-obese \ud groups. The thesis comprises three studies:\ud Classification of Aerobic and Gym-based Exercises from Accelerometer Output. This study \ud investigated whether accurate classification could be achieved from hip- or ankle-mounted \ud accelerometers for a programme of aerobic exercises and free-living activities. It also \ud examined whether accuracy was affected by obesity, and whether a single classifier could be \ud applied across BMI groups. The study achieved high classification accuracies (85% for hip \ud and 94% for ankle) for both obese and normal BMI groups using the same approach across \ud groups.\ud Walking Speed Estimation Using Accelerometry. This study aimed to develop a speed \ud estimation model that was applicable across BMI groups, and which utilised a hip-mounted \ud accelerometer. To achieve this, multiple accelerometer signal features were evaluated for use \ud in a linear speed estimation model, and performance was compared between obese and \ud normal BMI groups. The speed estimation algorithm achieved overall RMSE of 0.08ms-1\ud for \ud a mixed BMI group, which is comparable with previous research using homogeneous groups.\ud Prediction of Energy Expenditure from Accelerometer Output. This study aimed to identify \ud physiological and anthropometric parameters for use in an improved energy expenditure \ud estimation model. Model performance was tested on a mixed BMI group. The energy \ud expenditure prediction model incorporating subject attributes showed around 20% \ud improvement over the standard model. \ud This research found that current approaches to activity classification using accelerometry are \ud equally applicable to obese groups and normal BMI groups. Walking speed prediction was \ud shown to be possible from a hip-mounted accelerometer for both obese and normal BMI \ud groups. Energy expenditure estimation is improved by including subject-specific parameters \ud in the prediction model. Accelerometry is, therefore, a suitable tool for measuring different \ud aspects of physical activity for obese and mixed BMI groups.