Related Experiment Video
Updated: Aug 13, 2025
08:56
Clinical Assessment of Spatiotemporal Gait Parameters in Patients and Older Adults
Published on: November 7, 2014
14.0K
A Machine Learning Approach for Walking Classification in Elderly People with Gait Disorders.
Abdolrahman Peimankar1, Trine Straarup Winther1, Ali Ebrahimi1
1Centre of Health Informatics and Technology, The Mærsk Mc-Kinney Møller Institute, University of Southern Denmark, 5230 Odense, Denmark.
Sensors (Basel, Switzerland)
|January 21, 2023
Summary
This study developed a machine learning model to accurately classify walking activity in elderly individuals with mobility issues using back-worn accelerometers. The model shows promise for monitoring daily activities and improving care for those with dementia and Alzheimer's disease.
Area of Science:
- Biomedical Engineering
- Gerontology
- Machine Learning
Background:
- Elderly individuals with walking difficulties experience limited mobility, impacting their physical health and independence.
- Monitoring daily activities is crucial for improving the quality of life, especially for individuals with dementia and Alzheimer's disease.
- Sensor placement on the back is necessary for this population, posing challenges for activity detection.
Purpose of the Study:
- To develop and evaluate a Machine Learning (ML) based algorithm for accurately classifying walking activity in elderly individuals with walking difficulties.
- To address the challenges of sensor placement on the back for patients with dementia and Alzheimer's disease.
- To compare the performance of different ML classifiers for walking activity recognition.
Main Methods:
- Collected accelerometer data from elderly participants with walking difficulties.
- Extracted statistical, temporal, and spectral features from the time-series data.
- Utilized Particle Swarm Optimization (PSO) for feature selection.
- Trained and compared four ML classifiers (kNN, RF, Stack, XGB) using leave-one-group-out cross-validation (LOGO-CV).
Main Results:
- The Stacking Classifier (Stack) model achieved the highest performance.
- The Stack model demonstrated average sensitivity of 86.85%, positive predictive values (precision) of 93.25%, F1-score of 88.81%, and accuracy of 93.32% in classifying walking episodes.
- The ML models successfully classified walking episodes despite challenging back sensor placement.
Conclusions:
- The developed ML-based algorithm is effective in classifying walking activity in elderly individuals with walking disabilities.
- Back sensor placement is feasible and effective for monitoring daily activities in patients with dementia and Alzheimer's disease.
- This technology can significantly aid care home and rehabilitation center personnel in monitoring patient progress and improving quality of life.
