Predictive Modelling of Muscle Fatigue in Climbing

Sport climbing, a discipline demanding high levels of muscular strength, endurance, and cognitive planning, has gained consid erable popularity in recent years. The importance of managing muscle fatigue during climbing, which can substantially impair performance and potentially lead to injury, has not yet been thor oughly investigated. Predicting muscle fatigue can help to create tailored training programsandrefineclimbingstrategiestooptimise performance and safety. This study aimed to monitor and predict muscle fatigue during climbing. We conducted a multi-modal exper iment involving 20 climbers, measuring their muscle activity and fatigue via electromyography (EMG) and tracking their climbing trajectories through video recordings. We compared different linear autoregressive process (AR) methods and machine learning meth ods that predict muscle fatigue up to 5 seconds into the future. We also successfully proposed to extend the AR model to account for expected muscle fatigue as a function of distance climbed. While this extension improved prediction over the standard AR model, with a Root mean square error (RMSE) of 16.05 (±28.42%), the non linear Multilayer perceptron (MLP) and Gradient boosting (GB) models outperformed linear methods, with a lower RMSE of 15.09 (±27.94%)and 15.09 (±29.02%), respectively. Despite their lower ac curacy compared to non-linear models, the simplicity of calculating linear models could enable real-time predictions on wearable de vices, providing climbers with valuable, immediate feedback on muscle fatigue. This capability can significantly impact climbing research by providing practical tools for real-world applications, improving climbers' decision-making and enhancing safety and performance.