Javascript must be enabled for the correct page display

Predicting well-being with wearable sensor data

Akrum, Ivana (2019) Predicting well-being with wearable sensor data. Bachelor's Thesis, Artificial Intelligence.


Download (700kB) | Preview
[img] Text
Restricted to Registered users only

Download (144kB)


The world’s population is ageing rapidly but not more healthily. The healthcare industry needs some way to accommodate for the increasing healthcare needs of the elderly. This research proposes the idea of using classifiers in combination with wearable sensor data to monitor the well-being of the elderly. This will allow healthcare professionals to focus on those whom the algorithm has already identified as being unwell, thereby sparing them the time it would have taken to manually determine which patients need medical attention and which currently do not. After looking at literature on which factors influence health and cross-referencing that with which sensor data are feasible to measure within the scope of this research, the research aims to answer the question: Can classifiers be used to make predictions of well-being on the basis of sleep, physical activity, time away from home, and the circadian rhythm? To answer this question, an observational study of ten days was done for twenty participants in which a wearable measured the necessary health data. The participant well-being was assessed in three ways: through diagnoses from health professionals, through participant self-report, and through the sensor data itself. A K-Nearest Neighbour (KNN) and Random Forest (RF) classifier were trained that used the sensor data to make predictions of well-being as assessed through the data. The RF classifier gave the best results and showed an accuracy of 89%. However, the well-being assessed through sensor data was ultimately found to be unrepresentative of well-being, and the RF classifier performance was considerably worse when compared against the well-being assessed through health professionals and through self-report. As such, even the optimal classifier designed in this research cannot be used to make reliable predictions of well-being on the basis of sleep, physical activity, time away from home, and the circadian rhythm.

Item Type: Thesis (Bachelor's Thesis)
Supervisor nameSupervisor E mail
Vugt, M.K.
Degree programme: Artificial Intelligence
Thesis type: Bachelor's Thesis
Language: English
Date Deposited: 13 Aug 2019
Last Modified: 15 Aug 2019 09:12

Actions (login required)

View Item View Item