Feature Selection and Predictors of Falls with Foot Force Sensors Using KNN-Based Algorithms

Area of Science:

• Gerontology
• Biomechanics
• Medical Engineering

Background:

• Aging often leads to decreased lower extremity function, impacting daily life quality and increasing fall risk in the elderly.
• Predicting falls is crucial for proactive interventions and maintaining independence in older adults.

Purpose of the Study:

• To determine if objective physical function measures can predict future falls in the elderly.
• To develop a classification method for identifying elderly individuals at risk of falling.

Main Methods:

• Collected ground reaction force (GRF) data using foot force sensors, quantifying it with sample entropy.
• Utilized a feature selection algorithm and three k-nearest neighbor (KNN) classifiers (LMKNN, PNN, LMPNN) to classify subjects into fall-risk groups.
• Compared classification performance using functional movement tests (walking, sit-to-stand).

Main Results:

• The local mean pseudo nearest neighbor (LMPNN) classifier achieved 100% sensitivity, specificity, and accuracy in predicting falls.
• A subset of GRF features showed significant differences between fallers and non-fallers, supporting the classification findings.
• The developed method demonstrates high efficacy in distinguishing between elderly individuals at risk and not at risk of falling.

Conclusions:

• Objective GRF analysis, particularly with the LMPNN classifier, offers a highly accurate method for fall risk prediction in the elderly.
• This approach has the potential for use by non-experts to monitor balance and fall risk, improving elderly care.
• Accurate fall risk assessment can lead to timely interventions, enhancing the safety and well-being of the aging population.