Automatic Recognition and Analysis of Balance Activity in Community-Dwelling Older Adults: Algorithm Validation

Area of Science:

• Gerontology
• Biomedical Engineering
• Data Science

Background:

• Clinical assessments for mobility and balance identify fall risks in older adults.
• Inertial sensors offer quantitative, cost-effective data for community-based mobility and balance assessments.
• Current sensor methods often rely on manual observation or specific motion features.

Purpose of the Study:

• To develop an automatic motion data analytics framework using inertial sensor data.
• To analyze balance activities in community-dwelling older adults.
• To classify individuals at high risk of falls.

Main Methods:

• Recruited 59 community-dwelling older adults (mean age 81.86 years).
• Collected data using a body-worn inertial measurement unit (accelerometer and gyroscope) at the L4 vertebra.
• Employed a convolutional long short-term memory (LSTM) neural network for motion detection, followed by one-class SVM, LDA, and k-NN for classification.

Main Results:

• Achieved mean accuracies of 87% (sit-to-stand), 86% (360° turn), and 89% (stand-to-sit) in motion detection.
• Classified abnormal balance activities with accuracies of 90% (sit-to-stand), 92% (360° turn), and 86% (stand-to-sit) using POMA-B criteria.
• Demonstrated high classification accuracy for fall risk using one-class SVM and k-NN.

Conclusions:

• The sensor-based framework enables efficient balance assessment with reduced human effort.
• Provides a time-effective method for identifying and preprocessing inertial signals.
• Offers a flexible solution for continuous health monitoring in the community, potentially reducing healthcare burdens.