Using supervised learning machine algorithm to identify future fallers based on gait patterns: A two-year...

Area of Science:

• Gerontology
• Biomedical Engineering
• Data Science

Background:

• Falls in the elderly pose significant health challenges.
• Current tools for predicting fall risk in independent older adults are limited.
• Gait parameters are known indicators of fall incidence.

Purpose of the Study:

• To develop a predictive tool for identifying future fallers among independent older adults.
• To apply a supervised learning algorithm to gait data for fall risk classification.
• To build a classification tree distinguishing subsequent fallers based on gait patterns.

Main Methods:

• A two-year longitudinal study included 105 independent older adults (>65 years) with no recent fall history.
• Gait parameters (speed, stride length, symmetry, regularity, toe clearance) were recorded under various walking conditions.
• A supervised machine learning algorithm (J48) was used to create a classification tree from the recorded data.

Main Results:

• A classification tree was developed using gait patterns, gender, and stiffness.
• The model achieved 84% accuracy, correctly identifying 80% of future fallers.
• Key performance metrics included 80% sensitivity and 87% specificity.

Conclusions:

• Gait parameters and clinical data can effectively identify future fallers in independent older adults.
• This study presents the first predictive tool based on identified gait parameters.
• Further validation is recommended, but the tool shows promise for clinical application and research.