Muscle Fatigue Detection Integrating EMG and PPG with Explainable Machine Learning Across Key Shoulder Muscles in Rotational Movements

Muscle fatigue prediction is critical for improving performance and safety, especially in sports science and rehabilitation. This paper presents a comparative analysis of various machine learning models for predicting muscle fatigue using Electromyography (EMG) and Photoplethysmogram (PPG) data. Several machine learning algorithms, including Random Forest, SVM, Logistic Regression, Decision Trees, and other models, are evaluated for their ability to predict muscle fatigue. Some models are optimized through Grid Search for hyperparameter tuning. SMOTE-ENN (Synthetic Minority Oversampling Technique (SMOTE) with Edited Nearest Neighbors (ENN)) is used as a technique for class imbalance mitigation. Evaluations are made by metrics such as accuracy, precision, recall, and F1-score. Further, SHAP analysis, an Explainable AI method, was performed on the KNN model tuned through GridSearch, disclosing the top contributing features for muscle fatigue prediction in internal and external shoulder rotations; notably, the median frequency of the upper trapezius emerged as a key indicator of fatigue in internal rotation. The optimized KNN model achieved accuracies of 95.14% and 95.45% for internal and external rotation, respectively.