A framework for semi-supervised adaptive learning for activity recognition in healthcare applications

Gupta, Prankit; Caleb-Solly, Praminda

Abstract

With the growing popularity of the Internet of Things and connected home products, potential healthcare applications in a smart-home context for assisted living are becoming increasingly apparent. However, challenges in performing real-time human activity recognition (HAR) from unlabelled data and adapting to changing user health remain a major barrier to the practicality of such applications. This paper aims to address these issues by proposing a semi-supervised adaptive HAR system which combines offline and online recognition techniques to provide intelligent real-time support for frequently repeated user activities. The viability of this approach is evaluated by pilot testing it on data from the Aruba CASAS dataset, and additional pilot data collected in the Bristol Robotics Lab’s Assisted Living Studio. The results show that 71% of activity instances were discovered, with an F1-score of 0.93 for the repeating “Meal_Prep” activities. Furthermore, real-time recognition on the collected pilot data occurred near the beginning of the activity 64% of the time and at the halfway point in the activity 96% of the time.

• Presentation Conference Type: Conference Paper (published)
• Conference Name: 19th International Conference, EANN 2018
• Start Date: Sep 3, 2018
• End Date: Sep 5, 2018
• Acceptance Date: Jun 4, 2018
• Online Publication Date: Jul 27, 2018
• Publication Date: 2018
• Deposit Date: Jan 22, 2020
• Publicly Available Date: Apr 20, 2021
• Publisher: Springer Publishing Company
• Volume: 893
• Pages: 3-15
• Series Title: Communications in Computer and Information Science
• Series ISSN: 1865-0929
• Book Title: Engineering Applications of Neural Networks
• ISBN: 9783319982038
• DOI: https://doi.org/10.1007/978-3-319-98204-5_1
• Keywords: Activity recognition, Smart-home, Healthcare
• Public URL: https://uwe-repository.worktribe.com/output/4763358