Early Detection of Physical Fatigue via Multimodal Biomedical Signals with Supervised Learning Algorithms

Physical fatigue poses significant risks in safety-critical environments and athletic performance. This study presents a machine learning-based framework for early fatigue detection using physiological signals electromyography (EMG), oxygen saturation (SpO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf>), and heart rate (HR). In this study, the dataset which was titled Fatigue DetectS.csv, was used containing EMG, SpO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf>, and HR measurements and binary fatigue labels. $80 \%$ of the dataset is divided as training data and $20 \%$ as testing data. After feature normalization, five high-performance models are evaluated: Random Forest, Support Vector Machine (RBF), XGBoost, LightGBM and Multi-Layer Perceptron (MLP). All models were implemented using Python-based libraries and validated via $\mathbf{1 0}$-fold crossvalidation. Performance metrics included accuracy, F1 Score, Precision, Recall, ROC AUC, and computational time. Experimental results showed that RBF, SVM and MLP achieved the highest classification metrics (F1 Score = 0.94, Accuracy = 94%), while XGBoost and LightGBM offered superior computational efficiency with training times below 0.2 seconds. The dataset’s balanced class distribution ensured unbiased evaluation, and all models exhibited robust performance without overfitting. This comparative analysis highlights the feasibility of integrating machine learning with biomedical signals for fatigue monitoring in real time applications. While deep learning models like MLP provide high classification quality, tree-based models are preferable for low-latency systems such as wearables. The findings contribute to the development of intelligent health monitoring systems capable of reliable fatigue prediction in dynamic environments.