Aims and Objectives: The WEALTH project aims to develop machine learning methods to accurately process data from research grade and commercially available wearable accelerometer devices so that they can measure the daily time spent in the Physical Behaviours (PB: including physical activity, sedentary time and sleep). A second aim of the project is to test the feasibility of linking machine learning generated data from the accelerometers with data obtained using Ecological Momentary Assessment (EMA) in which where participants respond to questions through their smartphone. This will produce an integrated data collection system that will simultaneously capture PB and Eating Behaviour (EB) and their interrelation.

The objective of the “WEALTH” project is to develop standardised methods to derive daily physical and eating behaviours from wearable sensors and evaluate the interaction of both behaviours. To develop and validate machine learning models, 600 participants will be recruited to wear a range of state-of-the-art accelerometer devices including wearables for 7 days. Both physical and eating behaviours will be labelled using feedback from event-triggered and random EMA questions. This will lead to the development of an integrated data collection and processing infrastructure.

Results: WEALTH will simplify combined measurement of physical behaviours and diet. The ML methods will quantify daily time spent in PB from their distinctive acceleration patterns by identifying activities of daily living in the recorded data. Moreover, the software package developed will determine the energy expenditure of each activity and quantify the daily time spent at different exercise intensities and the daily energy expenditure. This can be important when evaluating achievement of WHO recommendations for PA in individuals and populations.

The EMA methods developed in WEALTH will provide a method of passively assessing the real-time impact of PB such as prolonged sedentary time and bouts of PA on eating behaviour. This is an important development, because existing tools generally measure the two separately with different sample times. The ability to target specific activities (for example, evening-time sedentary behaviour) for study, and to deliver EMA based questions during the behaviour will be important in future studies where the interaction of PB and EB is critical.

Expected Impact: The combination of ML processed data from wearable devices and smartphone based EMA’s will deliver a measurement solution linking PB and their context to EB. Tools provided by WEALTH will include those based on wearables and fitness trackers providing low-cost wrist-worn movement-sensors. The results will therefore have applications in research and management of data on PB and EB, in surveillance and monitoring at population level of PB and EB and in public health interventions to promote healthy PB and related EB.