Lung Functions and Self-Reported Physical Activity Level for the Classification of Physical Frailty in Older Adults Using a Machine Learning Approach

Introduction: Assessing physical frailty often requires multiple performance-based measures, which may be impractical for older adults with health impairments or in time-constrained settings. Lung function is a key predictor of frailty progression, and physical activity is strongly associated with lung function. However, whether these two factors alone can accurately classify frailty using machine learning (ML) remains unclear. This study evaluates the classification of frailty status using lung function and self-reported physical activity levels, comparing Neural Networks (NN) and Support Vector Machines (SVM). Methods: Data from 37 older adults (66.24 ± 4.34 years) were collected completely at community centers. Physical activity was assessed using the International Physical Activity Questionnaire (IPAQ), and lung function was measured via spirometry (Spirobank II, MIR, Italy). The dataset was split into training and testing with a ratio of 70:30. Within the training set, 10-fold cross-validation was applied to optimize model performance before final testing. Results: The NN model outperformed SVM, achieving 90.9%testing accuracy with high Area Under the ROC Curve (AUC) scores: frail (1.00), pre-frail (0.99), and robust (1.00). The Linear SVM model showed slightly lower discriminatory power. Conclusion: Lung function and physical activity levels effectively predict frailty status. The superior performance of the NN model supports its potential for early frailty detection. Future research should explore larger datasets and additional physiological factors to enhance predictive accuracy and clinical utility.