Optimized Photoplethysmography-Based Classification of Calf Muscle Fatigue Using Particle Swarm Optimization with Logistic Regression

This study investigates photoplethysmography (PPG) as a non-invasive, cost-effective alternative for real-time muscle fatigue monitoring, addressing limitations inherent to conventional methods like electromyography (EMG) and blood lactate testing. A PPG-based system was developed to classify fatigued versus non-fatigued states of the calf muscle using a DFRobot SEN0203 sensor at a 1000 Hz sampling rate. The raw PPG signals were segmented into 1-second intervals and processed to compute first and second derivatives—yielding vascular (VPG) and arterial (APG) photoplethysmograms—which enabled extraction of key features including heart rate (HR), heart rate variability (HRV), peak systolic and diastolic voltages, maximum systolic slope (u), minimum diastolic slope (v), and arterial stiffness indicators (b–a and c–a ratios). A Particle Swarm Optimization (PSO) algorithm was employed to optimize both feature selection and hyperparameters within a Logistic Regression (LR) model, achieving perfect classification accuracy (1.0) with training and prediction times of 0.0053 s and 0.0016 s, respectively. Notably, HRV and the minimum diastolic slope—reflecting autonomic regulation and vascular compliance—emerged as the most influential features with weights of 12.3747 and 23.9367. Comparative analyses revealed that although LightGBM matched the PSO-LR accuracy, neural network approaches performed poorly (0.50 accuracy), likely due to overfitting and limited training data. These findings underscore the viability of PPG for muscle fatigue monitoring, with promising applications in sports science, rehabilitation, and occupational health.