Identifying neuromuscular and mental fatigue in elite youth table tennis players using machine learning.

Performance in table tennis is known to be affected by both mental and physical fatigue. However, accurately predicting a player's fatigue state remains challenging, particularly when using non-intrusive measurement protocols. This study proposes a methodology for detecting fatigue based on the analysis of players' movement patterns following controlled fatigue induction in young elite table tennis players. Player movements were recorded using an instrumented racket equipped with an accelerometer and pressure sensors. Mental and physical fatigue were induced using established protocols and were validated through significant changes in reference markers, including the Rating of Perceived Exertion for mental fatigue (p < 0.05) and Maximum Voluntary Contraction for physical fatigue (p < 0.05). The resulting labelled datasets were used to train supervised machine learning models for both binary fatigue detection and multiclass fatigue classification. Among the evaluated models, K-Nearest Neighbors and Random Forest achieved the best performance, with recognition rates of approximately 84% for binary classification and 82% for multiclass classification. These results demonstrate that combining formal fatigue induction with instrumented movement analysis enables the creation of reliable datasets for predicting players' fatigue states from their gestures.