Detecting sleep deprivation from running biomechanics using machine learning classification: a comparison between wearable and laboratory motion capture

Sleep deprivation is associated with impaired endurance performance and an increased risk of running-related injury. Previous research has identified alterations in running biomechanics following a single night of sleep deprivation under laboratory conditions. However, whether these biomechanical changes can be detected using wearable technology remains unknown. Twenty-one recreationally active runners completed submaximal treadmill running under both normal sleep and total sleep deprivation conditions in a randomized crossover design. Biomechanical features were extracted simultaneously using a full-body motion capture system and a trunk-mounted wearable sensor. Five machine learning classifiers were evaluated in two classification tasks: a within-subject task using paired recordings from the same individual, and a between-subject task performed without individual baseline data. Within-subject classification exceeded chance level (label-permutation testing, p < 0.05 ) for most models and both measurement systems, with best accuracies of 84% for the wearable sensor (Logistic Regression) and 81% for the motion capture system (Random Forest). These findings indicate that sleep deprivation produces a systematic and individually consistent biomechanical signature during running. In contrast, between-subject classification failed across nearly all models and systems, with accuracies remaining close to chance level (˜50%), suggesting that inter-individual variability obscures the sleep-deprivation signal in the absence of personalized baseline data. Both systems converged on temporal organization, loading-related variables, and stride-to-stride variability as the most discriminative feature domains. Contrary to expectations, the laboratory motion capture system did not outperform the wearable sensor. Together, these findings suggest that individualized, baseline-referenced monitoring may be important for detecting sleep-deprivation-related changes in running gait, and suggest that a single trunk-mounted wearable sensor may provide a practical solution for real-world monitoring when paired recordings are available.