Abstract
BACKGROUND: Muscle fatigue represents a primary manifestation of exercise-induced fatigue. Electromyography (EMG) serves as an effective tool for monitoring muscle activity, with EMG signal analysis playing a crucial role in assessing muscle fatigue. This paper introduces a machine learning approach to classify EMG signals for the automatic detection of muscle fatigue. METHODS: Ten adult participants performed isometric contractions of lower limb muscles. The EMG signals were decomposed into multiple intrinsic mode functions (IMFs) using improved complementary ensemble empirical mode decomposition adaptive noise (ICEEMDAN). Time-domain, frequency-domain, time-frequency domain, and nonlinear features associated with muscle fatigue during isometric contraction were analyzed through EMG signals. Dimensionality reduction was achieved using t-distributed stochastic neighbor embedding (t-SNE), followed by machine learning-based classification of fatigue levels. RESULTS: The findings indicated that EMG signal characteristics changed significantly with increasing fatigue. The combination of support vector machines (SVM) and ICEEMDAN achieved an impressive accuracy of 99.8%. CONCLUSION: The classification performance of this study surpasses that of existing state-of-the-art methods for detecting exercise-induced fatigue. Therefore, the proposed strategy is both valid and effective for supporting the detection of muscle fatigue in training, rehabilitation, and occupational settings.