Identifying Fallers Based on Functional Parameters: A Machine Learning Approach

Abstract

Falls are a leading cause of fracture and mortality in older adults, and hence represent a considerable socioeconomic burden in aging societies. Detection of individuals at a high risk of falls and evaluation of associated factors enable implementation of targeted therapies and timely intervention. The most common indicator for fall prediction is history of falling, but this is a subjective predictor and fails to detect first-time fallers simply because it is absent in such cases. In this study, we used functional variables extracted from multiple functional domains, and implemented several machine learning (ML) methods to classify fallers vs non-fallers retrospectively. We also performed feature importance analysis to provide an insight into the underlying features. Performed within a cross-validation setting, we identified the ML algorithm that best maps individuals’ functional measures to their fall status. In addition, we applied this algorithm for prospective identification of fall risk. In retrospective classification, k-nearest neighbours (KNN) model achieved a sensitivity of 74% and a specificity of 75%. In prospective evaluation, it achieved sensitivity and specificity of 80%. These results reflect the superior capability of machine learning in fallers identification even with a very small dataset.