Identification of Selective Attention Strategies in Older Adults Using Biomechanical Analysis of Gait Kinematic Data Based on Machine Learning

Abstract

Background: Aging is associated with gradual declines in neuromuscular and skeletal function, which affect motor and cognitive control and alter gait patterns. Selective attention, particularly the distinction between internal and external focus, plays a critical role in motor control in older adults. Objectives: This study aimed to evaluate the ability of machine learning models using joint kinematic features to distinguish between two attentional strategies during gait in older adults. Methods: Nineteen healthy older women participated in this study. Kinematic data for 12 degrees of freedom of lower- and upper-limb joints in the sagittal plane were recorded using the OpenCap system. Each participant walked along a 5-meter path under two attentional conditions: internal focus and external focus. Angular displacement, angular velocity, and combined features were used to train bagged decision tree models. The first three principal components were used for dimensionality reduction, and the models were evaluated on test data. Results: The angular displacement-based model achieved the highest test accuracy of 76.2% and a notable area under the receiver operating characteristic curve, whereas the angular velocity and combined models showed lower performance, with approximately 62% accuracy. Under the current summary-statistics feature framework, angular displacement features demonstrated stronger discriminative capability, likely reflecting more stable and regular gait cycles in older adults. Conclusions: Joint angular displacement features offer practical and reliable indicators for identifying attention strategies in older adults. These findings support the development of low-cost, noninvasive tools for cognitive-motor monitoring, rehabilitation, and fall prevention. Future research with larger samples and time-series models may improve generalizability and predictive accuracy.