Identification Of Physical Fatigue In Interval Training: A Machine Learning Approach

Competitive running has a positive influence on fitness and psychology of runners. However, demanding training sessions can lead to lower limb injuries, especially in interval training. To help coaches choose ideal training loads, physical fatigue should be monitored. PURPOSE: To detect physical fatigue in the lower limbs and trunk during interval training using biomechanical parameters. METHODS: 5 experienced runners (M, 23 ± 6 years, 180 ± 10 cm, 67 ± 5 kg) ran 8 laps of 400 m on an athletic track at a constant speed (18.5 ± 0.7 km/h) with breaks (recovery time 40 ± 1 seconds) between laps. They wore 8 IMUs (240 Hz) attached to their body (feet, tibias, thighs, pelvis and sternum). Gait cycles were segmented based on peak downwards velocity of the pelvis. A random forest classifier was trained with features derived from segment accelerations and joint angles and was used to distinguish strides between Lap 1 (non-fatigued) and Lap 8 (fatigued). Classifier performance is assessed by means of LOSO cross-validation. RESULTS: Perceived exertion increased between Lap 1 (avg RPE = 9.4) and Lap 8 (avg RPE = 13.2). Fatigue was detected with an accuracy of 81% (sensitivity =76%, specificity = 86%). Changes in biomechanical parameters can be found in Table 1. CONCLUSIONS: Fatigue can be detected in interval training due to changes in biomechanical parameters. In previous research we were able to detect fatigue in a continuous run (speed = 10.6 ± 1.4 km/h) on an athletic track with an accuracy of 90%. Here we extended the validity of our method to trainings involving recovery periods for well-trained runners. Sensor technology and machine learning improve explicit identification of physical fatigue in specific body segments compared to traditional global fatigue estimates (i.e. heart rate, RPE). Supported by H2020 GA #826304