Uncertainty Investigation for Personalised Lifelogging Physical Activity Intensity Pattern Assessment with Mobile Devices

Abstract

Lifelogging physical activity (PA) assessment is crucial to healthcare technologies and studies for the purpose of treatments and interventions of chronic diseases. Traditional lifelogging PA monitoring is conducted in non-naturalistic settings by means of wearable devices or mobile phones such as fixed placements, controlled durations or dedicated sensors. Although they achieved satisfactory outcomes for healthcare studies, the practicability become the key issues. Recent advance of mobile devices make lifelogging PA tracking for healthy or unhealthy individuals possible. However, owning to diverse physical characteristics, immaturity of PA recognition techniques, different settings from manufactories and a majority of uncertainties in real life, the results of PA measurement is leading to be inapplicable for PA pattern detection in a long range, especially hardly exploited in the wellbeing monitoring or behaviour changes. This paper investigates and compares uncertainties of existing mobile devices for individual's PA tracking. Irregular uncertainties (IU) are firstly removed by exploiting Ellipse fitting model, and then monthly density maps that contain regular uncertainties (RU) are constructed based on metabolic equivalents (METs) of different activity types. Five months of four subjects PA intensity changes using the mobile app tracker Moves [1] and Google Fit app on wearable device Samsung wear S2 are carried out from a mobile personalised healthcare platform MHA [2]. The result indicates that uncertainty of PA intensity monitored by mobile phone is 90% lower than wearable device, where the datasets tend to be further explored by healthcare/fitness studies. Whilst PA activity monitoring by mobile phone is still a challenging issue by far due to much more uncertainties than wearable devices.