Fall risk classification with posturographic parameters in community-dwelling older adults: a machine learning and...

Area of Science:

• Gerontology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Computerized posturography in standing positions is used to assess fall risk in older adults and specific disease groups.
• Machine learning (ML) offers advantages over traditional regression for analyzing complex, high-dimensional, non-linear, and correlated posturographic data.
• Explainable Artificial Intelligence (XAI) is crucial for increasing the interpretability of ML models in fall risk assessment.

Purpose of the Study:

• To employ ML algorithms for classifying fall risks in community-dwelling older adults.
• To enhance the interpretability of these ML models using an XAI approach.
• To investigate the efficacy of posturographic parameters in predicting fall risk.

Main Methods:

• Analysis of 215 participants' personal metrics and posturographic data from four standing postures.
• Utilized two classification criteria: previous fall history and the Timed-Up-and-Go (TUG) test scores.
• Employed meta-heuristic methods for feature selection and the SHapley Additive exPlanations (SHAP) method for model interpretability.

Main Results:

• Posturographic parameters effectively classified participants based on TUG scores (>10s vs. <10s), but were less effective for classification by previous fall history.
• Feature selection improved model accuracy when using TUG scores, with the Slime Mould Algorithm showing the best performance (accuracy: 0.72-0.77, AUC: 0.80-0.90).
• SHAP values successfully illustrated the importance of different features within the classification models.

Conclusions:

• Standing posturography parameters can accurately classify fall risk in older adults, particularly when correlated with TUG scores.
• Feature selection enhances the performance of ML models for fall risk classification.
• The integration of ML and XAI holds significant potential for developing more accurate and reliable fall risk assessment tools.