Classificating middle-aged and older adults through physiological and functional measures

Abstract

Aging affects the functional capacity of individuals by causing gradual changes in metabolic, gait, balance and muscle functions. Identifying these changes between middle-aged (45–64) and older (≥65) adults is critical to understanding the biological and functional effects of aging. This study aims to evaluate the differences between middle-aged and older adults in an objective and scalable manner by analyzing metabolic indicators, gait parameters, balance measurements and muscle functions using machine learning (ML) methods. In this study, 57 high-dimensional variables from the MIDUS dataset including gait parameters (e.g. gait speed, cadence, cycle time), muscle function, balance measurements (e.g. path length, swing area), bone mineral density and bioelectrical impedance spectroscopy markers were used. Supervised ML models were applied to classify the age groups: Partial Least Squares Discriminant Analysis (PLS-DA), Principal Component Analysis-Linear Discriminant Analysis (PCA-LDA), Support Vector Machine (SVM), and k-Nearest Neighbors (k-NN). Venetian blind cross-validation approach was applied to evaluate the model performance. Among the models, SVM showed the highest classification accuracy (87 %) on the training data and 77 % accuracy on the testing data. PLS-DA model achieved 82 % accuracy in training and 86 % in testing. While k-NN model showed 87 % accuracy in training, it dropped to 68 % in testing. In terms of sensitivity and specificity values, SVM showed the best performance (96 % sensitivity, 67 % specificity - training; 86 % sensitivity, 55 % specificity - test), while PLS-DA and PCA-LDA models exhibited similar trends. The results show that walking speed, cadence, and balance measurements provide significant contributions to age group discrimination. These findings highlight the role of neuromuscular and physiological factors in functional decline due to aging, demonstrating the potential of machine learning-based classification in aging research.