Identification of fatigue-related kinematic changes in elite runners using a support vector machine approach

Understanding the kinematic changes underlying fatigue is essential in running biomechanics. The aim of this study was to identify fatigue-related kinematic changes in elite runners using a support vector machine approach. Full-body kinematics of thirteen trained runners were recorded in a non-fatigued and a fatigued state during treadmill running at their individual fatigue-speed. A support vector machine was trained and used to identify kinematic differences between the non-fatigued and fatigued state based on principal component scores. Strides during non-fatigued and fatigued running could be separated with 99.4% classification accuracy. Four upper limb (two shoulder and two elbow), four lower limb (one ankle, two knee and one hip) and two trunk (one thoracic and one lumbar spine) principal component scores were identified as most discriminative kinematic features between non-fatigued and fatigued running. The findings of the study suggest the feasibility of a support vector machine approach to identify subtle fatigue-related kinematic changes in elite runners.