Inertial sensor-based gait classification for frailty status in older adults: A cross-sectional study.

Area of Science:

• Gerontology
• Biomedical Engineering
• Data Science

Background:

• Frailty in older adults is linked to functional declines and unstable gait.
• Objective assessment of gait is crucial for identifying frailty.

Purpose of the Study:

• To develop and evaluate a gait-based system for detecting frailty in older adults using wearable sensors.
• To identify optimal gait features for accurate frailty classification.

Main Methods:

• Gait data (acceleration, angular velocity) were collected from frail and non-frail older adults using a wireless tri-axial inertial measurement unit (IMU).
• Signal processing included Savitzky-Golay and Butterworth filtering to remove noise.
• Statistical analysis identified significant gait features, and machine learning models (k-NN, SVM, Random Forest) were employed for classification.

Main Results:

• Initial models achieved 84-89% accuracy in classifying frailty based on gait features.
• An optimized feature extraction scheme significantly improved classification metrics to over 95% accuracy.
• The final system achieved 96% accuracy using only one minute of data from a portable, low-cost IMU system.

Conclusions:

• IMU-based gait analysis provides an objective and accurate method for frailty classification in older adults.
• The proposed optimal feature extraction scheme enhances system performance, offering a scalable solution for community-based frailty detection.
• Wearable sensor technology holds significant potential for improving geriatric health assessments and supporting aging-in-place strategies.