EMG-Based Multi-Class Gesture Recognition with Normalized Muscle Power Evaluation

The analysis of musculoskeletal system movements using electromyography (EMG) signals is a fundamental requirement in fields such as prosthetic control, human-machine interaction, and neuromuscular rehabilitation. This study presents a comprehensive approach that not only evaluates movement recognition accuracy but also quantitatively assesses the level of muscle force required for each movement. In the study, the muscle loading profile of each hand movement was created using EMG signal energy normalized to the Rest state. Five different classifier models were compared under 5-fold cross-validation (CV) and Leave-One-Subject-Out (LOSO) protocols. The results showed that the Extension movement had the highest normalized power value and that classification accuracy reached its highest level with SVM-RBF (86.95%). Furthermore, Out-of-Bag (OOB) error analysis revealed that the model converged stably around 600–800 trees, while accuracy differences between individuals were attributed to physiological variations. The proposed framework offers a new evaluation perspective for both ergonomic task design and clinical performance monitoring by assessing gesture recognition performance alongside muscle strength requirements.