Fatigue Detection in Running with Inertial Measurement Unit and Machine Learning

To date athlete/patient fatigue has been assessed using expensive laboratory equipment. Inertial measurement unit (IMU) offer an opportunity to provide low-cost and non-intrusive fatigue assessment. The aim of this study was to determine if in combination or in isolation, IMUs positioned on the low extremities are capable of distinguishing between fatigued and un-fatigued running states and to predict the degree of fatigue. A running fatigue dataset based on multiple IMUs was constructed by recording inertial data during running to a state of fatigue. In addition to the inertial data from the IMUs, the perceived level of exertion was monitored for each participant as an indication of their physical fatigue level. Random forest (RF) and support vector machine (SVM) model validation was performed on the dataset to classify the running fatigue and fatigue levels. Classification effect of RF was better than SVM; the classification accuracy improved with the increase of sensors; the accuracy of tibial IMU data on RF accomplished 87.21%; the classification accuracy of combination of tibia and thigh IMUs was the highest at 91.10%. This study highlights the potential of inertial sensor to objectively estimate the level of fatigue during running by detecting minor deviations in lower extremity biomechanics.