Feature Selection and Validation of a Machine Learning-Based Lower Limb Risk Assessment Tool: A Feasibility Study.

Area of Science:

• Biomechanics
• Machine Learning
• Gerontology

Background:

• Early identification of locomotive degradation is crucial for preventing further decline.
• Traditional methods for assessing locomotive function often require significant manpower and infrastructure.
• Locomotive Syndrome (LS) is a significant concern in aging populations, impacting mobility and quality of life.

Purpose of the Study:

• To develop and evaluate machine learning models for lower limb skill assessment.
• To utilize readily available exercise features (squat and one-leg standing) as input for ML classifiers.
• To provide a non-labor-intensive and infrastructure-light alternative for detecting Locomotive Syndrome (LS).

Main Methods:

• Utilized nine squat and four one-leg standing exercise features as input parameters.
• Employed Artificial Neural Network (ANN) with two hidden layers and Rectified-Linear-Unit (ReLU) activation.
• Applied Random Forest (RF) regressor with varying numbers of estimators (5-100).
• Output layers were based on the Short Test Battery Locomotive Syndrome (STBLS) test, assessing sit-stand, 2-stride, and Geriatric Locomotive Function Scale (GLFS-25) scores.

Main Results:

• The ANN model achieved correlations of 0.59 for stand-up and 0.76 for 2-stride scores.
• The RF regressor yielded R-squared values of 0.86 for stand-up, 0.79 for 2-stride, and 0.73 for GLFS-25 scores.
• These results demonstrate the potential of ML in accurately predicting STBLS assessment scores.

Conclusions:

• Machine learning models, particularly RF, show high accuracy in predicting STBLS scores for locomotive function assessment.
• This ML-based approach offers a promising, resource-efficient method for early detection and management of Locomotive Syndrome (LS).
• The study highlights the feasibility of using simple exercise data for automated and objective locomotive skill evaluation.