Related Experiment Video
Updated: Mar 26, 2026
08:05
Design and Analysis for Fall Detection System Simplification
Published on: April 6, 2020
11.3K
The Effect of Personalization on Smartphone-Based Fall Detectors
Carlos Medrano1, Inmaculada Plaza2, Raúl Igual3
1EduQTech, E.U. Politécnica de Teruel, University of Zaragoza, c/Atarazana 2, 44003 Teruel, Spain. ctmedra@unizar.es.
Sensors (Basel, Switzerland)
|January 23, 2016
Summary
Personalizing smartphone fall detectors improves performance, especially for individuals with unique movement patterns. A personalized Nearest Neighbor (NN) detector achieved performance comparable to a non-personalized Support Vector Machine (SVM).
Area of Science:
- Biomedical Engineering
- Wearable Technology
- Machine Learning for Health
Background:
- High fall risk in elderly, Parkinson's patients, and neuro-rehabilitation.
- Need for robust fall detection systems for timely intervention.
- Limitations of non-personalized fall detection algorithms.
Purpose of the Study:
- To investigate the performance enhancement of personalized smartphone-based fall detectors.
- To compare personalized novelty detection algorithms with a traditional supervised algorithm.
- To assess the impact of individual-specific training data on fall detection accuracy.
Main Methods:
- Utilized a public dataset for fall detection research.
- Investigated four algorithms: Nearest Neighbor (NN), Local Outlier Factor (LOF), One-Class Support Vector Machine (OneClass-SVM), and Support Vector Machine (SVM).
- Evaluated personalization by comparing subject-specific training data against data from other subjects.
Main Results:
- Personalization generally increased fall detector performance across algorithms.
- Performance gains from personalization varied significantly between individuals.
- A personalized NN detector achieved an AUC of 0.9861, comparable to a non-personalized SVM (AUC 0.9795).
Conclusions:
- Personalized fall detection systems show potential for improved accuracy.
- The effectiveness of personalization is subject-dependent.
- Novelty detection algorithms, like NN, can achieve high performance with personalized training using daily living activities.
