Predicting risk of falling in older adults using supervised machine learning: a comparative analysis of model...

Area of Science:

• Gerontology
• Biomedical Engineering
• Data Science

Background:

• Falls in older adults represent a significant health concern.
• Developing reliable predictive models for fall risk is crucial for clinical relevance.
• This study investigates machine learning (ML) for fall risk prediction using functional measures.

Purpose of the Study:

• To evaluate the performance of supervised ML algorithms in predicting fall risk.
• To compare ML models against a logistic regression benchmark.
• To identify key functional predictors of falls in older adults.

Main Methods:

• Analyzed data from 94 older adults with comprehensive physical function and balance assessments.
• Implemented and compared four ML models: PLS-DA, LDA, SVM, and k-NN.
• Evaluated model performance using cross-validation, sensitivity, specificity, precision, accuracy, and AUC.

Main Results:

• All ML models demonstrated strong discriminatory power for fall risk.
• PLS-DA achieved the highest performance metrics (sensitivity, specificity, precision, accuracy, AUC).
• Logistic regression identified key predictors like 6-minute walk test (6MWT), 30-second chair stand test (30CST), and Berg balance scale (BBS).

Conclusions:

• Supervised ML models, particularly PLS-DA, effectively predict fall risk in older adults.
• ML models improve classification by integrating multidimensional patterns from functional measures.
• Findings classify fall history; further research is needed for definitive future fall prediction.