Faller Classification in Older Adults Using Wearable Sensors Based on Turn and Straight-Walking Accelerometer-Based...

Area of Science:

• Biomechanics
• Gerontology
• Wearable Technology

Background:

• Falls are a major health concern for the elderly, necessitating effective preventative strategies.
• Accurate faller classification is crucial for implementing timely interventions.
• Wearable sensors offer a promising avenue for objective fall risk assessment.

Purpose of the Study:

• To develop and evaluate a novel wearable-sensor based method for classifying elderly fallers.
• To investigate the utility of accelerometer-based gait features from straight walking and turns for faller classification.
• To identify optimal feature sets and machine learning models for accurate faller identification.

Main Methods:

• Seventy-six older adults (prospective fallers and non-fallers) participated in a six-minute walk test with leg and pelvis accelerometers.
• Gait data was segmented into straight walking and turn sections.
• Cross-validation was used to assess the performance of various classifier-model-feature selector combinations.

Main Results:

• The optimal model utilizing turn data, a random forest classifier, and a select-5-best feature selector achieved 73.4% accuracy.
• A refined feature subset, focusing on specific shank and lower back acceleration patterns during turns, improved classification to 77.3% accuracy.
• Classification performance metrics (accuracy, sensitivity, specificity, MCC) were consistently higher when using turn data compared to straight walking data.

Conclusions:

• Wearable sensor-based analysis of turning gait is effective for classifying elderly fallers.
• Specific accelerometer-derived gait features from turns hold significant potential for fall risk prediction.
• This technology can support the development of targeted preventative care strategies for older adults at risk of falling.