Home – Home – Jornals – Scientific journal “Vestnik NSUEM” 2026 number 2

This paper addresses the problem of detecting cognitive fatigue in online learning environments using eye-tracking data. The proposed approach combines statistical analysis (Pearson correlation and one-way ANOVA) with machine learning classification. Based on a publicly available dataset containing 2,784 observations with 41 eye-tracking parameters (including pupil diameter, fixation duration, time to first fixation, and peak saccade velocity), the study classifies fatigue levels into three categories: low, moderate, and high. Three classifiers are compared: Naïve Bayes, k-Nearest Neighbors, and Random Forest. The results show that Random Forest achieves the highest performance with 87 % accuracy, 86.6 % recall, and 86.9 % F1-score. A strong negative correlation (r = -0.74) is revealed between pupil diameter and fatigue level, indicating pupil constriction as a reliable fatigue marker. Limitations associated with remote webcam-based eye-tracking (lower sampling rate and reduced accuracy compared to lab-grade equipment) are discussed. The findings confirm the feasibility of real-time cognitive fatigue monitoring for adaptive e-learning systems that adjust instructional content based on the learner’s mental state.