Mental Fatigue Detection from ECG Data Using Active Machine Learning with a Query-by-Committee Strategy

This study explores the use of Active Machine Learning (AL) in the identification of mental fatigue from heart rate variability (HRV) data extracted from electrocardiograms (ECG). Based on the support vector machine (SVM) classifier, which demonstrated the best predictive performance in the reference study, we applied AL with the Query by Committee (QBC) strategy. This approach utilized a committee of models, including SVM, K-Nearest Neighbors (KNN), and Random Forest, to identify informative samples, significantly reducing the need for large volumes of labeled data. The method was evaluated with data collected during cognitive task performance and resting periods, considering time windows of 3, 4, and 5 minutes and batches of unlabeled data of sizes 20, 40, and 60. With a batch size of 20 and a 5-minute window, the model achieved an AUC of 0.837 and an accuracy of 75%, results comparable to those obtained in the reference study. This work demonstrates that AL is an effective strategy for optimizing mental fatigue classification, even in resource-constrained scenarios, enabling high-quality results with reduced datasets.