IRMA-International.org: Creator of Knowledge
Information Resources Management Association
Advancing the Concepts & Practices of Information Resources Management in Modern Organizations

Smartwatch-Based Semi-Supervised Fall Detection Using Anomaly Detection Techniques

Smartwatch-Based Semi-Supervised Fall Detection Using Anomaly Detection Techniques
View Sample PDF
Author(s): Meshall Alshalaan (King Abdullah University of Science and Technology, Saudi Arabia), Fouzi Harrou (King Abdullah University of Science and Technology, Saudi Arabia)and Ying Sun (King Abdullah University of Science and Technology, Saudi Arabia)
Copyright: 2026
Pages: 34
Source title: Improving Quality of Life for People with Disabilities Through Smart Technologies
Source Author(s)/Editor(s): Ikram Ur Rehman (University of West London, UK), Moustafa Nasralla (Prince Sultan University, Saudi Arabia), Drishty Sobnath (Heriot Watt University, UAE), Muazzam Ali Khan Khattak (Quaid-i-Azam University, Pakistan)and Sundus Ali (NED University of Engineering and Technology, Pakistan)
DOI: 10.4018/979-8-3373-2033-5.ch006

Purchase

View Smartwatch-Based Semi-Supervised Fall Detection Using Anomaly Detection Techniques on the publisher's website for pricing and purchasing information.

Abstract

Fall detection systems are crucial for seniors to mitigate significant health risks and severe fall-related injuries. Traditional supervised methods rely on extensive labeled datasets, which are challenging and time-consuming. Imbalanced datasets, where fall events are rare compared to daily activities, further complicate training. To bypass these challenges, we propose a semi-supervised fall detection approach using a One-Class Support Vector Machine (1SVM) with accelerometric data from smartwatches. 1SVM needs only fall-free data for training and effectively detects falls without labels. Specifically, we applied 1SVM to six features extracted from the accelerometric data, smoothed magnitude, moving window min-max difference, variance, maximum, and minimum. We evaluated this approach using publicly available smartwatch data named the SmartFall dataset, demonstrating 1SVM's superior performance with an accuracy of 87%, outperforming SVM, k-nearest neighbors (KNN), and Naive Bayes.

Related Content

Yogita Lamba, S. Srinivasan, Ajay Kumar Singh. © 2026. 44 pages.
Nthabiseng Istorina I. Mahetlana, Marubini Christinah Sadiki. © 2026. 30 pages.
J. John Shiny, S. Haranya, T. Sathiyarupa, Jaithun Shifaya B. S., P. Y. Sivanithi. © 2026. 44 pages.
Ghulam Fiza, Hira Mariam. © 2026. 48 pages.
Soorya Sathish, Cristina Turcanu. © 2026. 36 pages.
Meshall Alshalaan, Fouzi Harrou, Ying Sun. © 2026. 34 pages.
Ajay Menon, Mahmoud A. A. Mousa. © 2026. 36 pages.
Body Bottom