Comparative Analysis by Machine Learning of Geriatric Frailty and Alzheimer's Disease Classification Using...

Frailty syndrome and Alzheimer's disease are prevalent conditions in the elderly that are associated with aging, decreased quality of life, and a significant healthcare burden. Evidence for a relationship between physical frailty and neurodegenerative decline is accumulating. This study analyzed two independent datasets, a frailty dataset based on gait and mobility parameters and an AD dataset with clinical, functional and lifestyle variables, in order to evaluate and compare their classification performance using machine learning. Features were optimized using dimensionality reduction techniques to keep predictors of clinical significance and hyperparameter optimized Random Forest models were built to develop the best model. Evaluation was performed with Accuracy, F1-score, Matthews Correlation Coefficient and Area Under the Curve. The results showed that the models constructed on the whole AD dataset achieved maximum predictive power with an accuracy of 0.946, which was slightly increased to an accuracy of 0.948 after the selection of significant features. Diagnostic models based on frailty were able to demonstrate an ACC predictive capacity of 0.6418, and in terms of feature selection, improvements appeared in all indicators. Regarding the features derived from Alzheimer's disease associated with geriatric frailty, they managed to surpass the ACC frailty features of 0.741 alone, suggesting some intercalation mechanisms between neurodegeneration and physical vulnerability. These findings show that machine learning algorithms accompanied by feature selection improve clinical discrimination and prediction of frailty and neurodegenerative disorders, which offers a promising aspect for geriatric assessment. The frailty models analyzed demonstrated an ACC predictive capacity of 0.6418, even though feature selection improved all indicators. Alzheimer's disease-derived features associated with frailty outperformed features in the frailty dataset with an ACC of 0.741, suggesting the mechanism of overlap between neurodegeneration and physical vulnerability. These results support the theory of a motor-cognitive aging continuum, indicating that algorithmic machine learning techniques coupled with feature selection mainly provide computational validation for the biological intersection of neurodegeneration and physical frailty, rather than forming an independent predictive clinical model. Using these algorithms the study highlights shared pathophysiological mechanisms, providing a significant insight into systemic geriatric deterioration.