Fatigue prediction from sensor data using functional data analysis

John D. Andrew, University of Galway

Co-authors: Brian Caulfield, University College Dublin; Andrew J. Simpkin, University of Galway

Abstract: Running while fatigued increases the risk of injury. Objective fatigue detection in a laboratory setting is expensive and cannot provide real-time feedback. Wearable sensors provide a compelling alternative as they can capture different running kinematics that could give real-time feedback. In this paper, we analysed sensor data from 19 participants who ran both while healthy and when fatigued. Each individual’s data included several hundred longitudinal functional strides while healthy and while fatigued. We used multilevel Functional Principal Component Analysis (mFPCA) to preprocess the data and generate important features which are then used to predict fatigue. A logistic regression model using mFPC scores as predictors performed well in predicting fatigued strides, achieving accuracy, sensitivity, specificity, and Area Under the Receiver Operating Characteristic curve (AUROC) of 72%, 74%, 70% and 84%, respectively. These results show the potential of using wearable sensor technology to prevent running-related injuries, improve performance, and improve prognosis and monitoring in other fields.