Related Experiment Video
Updated: Apr 14, 2026
06:49
Evaluation of a Smartphone-based Human Activity Recognition System in a Daily Living Environment
Published on: December 11, 2015
9.4K
Feature selection for wearable smartphone-based human activity recognition with able bodied, elderly, and stroke
Nicole A Capela1, Edward D Lemaire2, Natalie Baddour3
1Ottawa Hospital Research Institute, Ottawa, Canada; Department of Mechanical Engineering, University of Ottawa, Ottawa, Canada.
Plos One
|April 18, 2015
Summary
Human activity recognition (HAR) using smartphone sensors can monitor patient mobility. Classifier-independent feature selection methods improve HAR system accuracy across diverse populations, including elderly and stroke patients.
Area of Science:
- Biomedical Engineering
- Rehabilitation Technology
- Wearable Sensors
Background:
- Human activity recognition (HAR) using wearable sensors offers insights into patient mobility for rehabilitation.
- Smartphones provide a low-cost, minimally invasive method for mobility monitoring via accelerometer and gyroscope sensors.
- Current HAR systems often use customized datasets and feature extraction methods, predominantly from able-bodied individuals.
Purpose of the Study:
- To investigate classifier-independent feature selection for human activity recognition.
- To identify common and population-specific features for mobility tasks.
- To evaluate the effectiveness of selected feature subsets across different populations (able-bodied, elderly, stroke patients).
Main Methods:
- Collected smartphone accelerometer and gyroscope data from 44 participants across able-bodied, elderly, and stroke populations performing 41 mobility tasks.
- Calculated 76 signal features and selected subsets using Relief-F, Correlation-based Feature Selection, and Fast Correlation Based Filter methods.
- Evaluated feature subsets using Naïve Bayes, Support Vector Machine, and j48 Decision Tree classifiers.
Main Results:
- Identified common signal features applicable to all populations, with specific differences noted in the stroke patient subset.
- Feature subsets achieved similar or improved classification accuracy compared to using the entire feature set.
- Selected features demonstrated classifier-independent utility for HAR systems.
Conclusions:
- Classifier-independent feature selection is effective for improving HAR systems.
- The identified feature subsets can enhance HAR system development and performance across diverse populations.
- This approach facilitates more robust and adaptable mobility monitoring for rehabilitation specialists.
