Investigate how to design cost-effective wearable intelligence techniques with dynamic active learning algorithms

investigate how to design cost-effective wearable intelligence techniques with dynamic active learning algorithms to objective quantification and sensitive measure of personal physical activity (PA) behaviour in free-living environments.
For curbing the global prevalence of physical inactivity and the associated average of 5.3 million deaths per year, the importance of the Physical Activity (PA) research has been demonstrated by World Health Organisation (WHO) reports. UK estimates in 2017 suggest over a quarter of people aged 16 years and over are categorised as physically inactive'. Research has shown that the prevalence of wearable devices and smartphones enable people to track and manage daily PA effortlessly, and potentially improve their health outcomes. But these wearable technologies suffer from low accuracy and weak robustness of objectively qualifying PA in free-living environments due to shortage of cost-effective sensors, unstandardised baseline dataset, effective and efficient learning algorithms, etc. Technological developments in the academic community will be examined through significant joint research and development between the University of Sheffield and Ant-Data Ltd that will examine how to seamlessly integrate advanced machine learning techniques into wearable systems for objective qualification of PA behaviour.

The research aims to result in a real-life wearable PA baseline dataset, machine learning models and aligned journal papers. The main aim is to investigate how to design cost-effective wearable intelligence techniques with dynamic active learning algorithms to objective quantification and sensitive measure of personal physical activity (PA) behaviour in free-living environments. This project addresses three key areas: i) Lack of holistic investigations on establishing standardised PA behaviour baseline datasets using cost-effective wearable technologies that collect personal PA data in free-living environments; ii) Lack of effective feature selection techniques for extracting reliable features from unlabelled and uncertain PA data, and iii) Lack of useful machine learning techniques for improving accuracy and robustness of qualification of PA behaviour.