Detecting mental fatigue in online learners using EEG-based emotional metrics and ensemble learning

Abstract

Mental fatigue is a brain state that causes energy levels to decline due to engagement in a prolonged period of cognitive tasks, such as online learning. It results attention deficits and poor academic performance by reducing focus and engagement during online learning sessions. It is essential to address mental fatigue early because it leads to permanent cognitive deficits and emotional impairments. Existing subjective assessments like surveys, interviews and self-report questionnaires are time-consuming and often reported as bias, since humans cannot accurately measure their cognitive performance. Although there are numerous machine learning (ML) approaches for mental fatigue detection using electroencephalogram (EEG) signals, there is a lack of study on mental fatigue detection among online learners. Therefore, this study aims to use EEG-derived emotional states metrics for mental fatigue detection by identifying existing EEG-based emotional metrics and mental fatigue detection techniques through literature review, then developing and evaluating machine learning models. Thus, this study employs Emotiv Performance Metrics (EPM) including engagement, excitement, interest, stress and relaxation. The mental fatigue detection consists of EPM metrics collection using an Emotiv Insight headset, and ML model development including Logistic Regression (LR), Support Vector Machine (SVM), Multilayer Perceptron (MLP) and Ensemble Learning model. A total of 10 students participated in the data collection session. However, data from only 8 participants were included in the mental fatigue detection. The two participants were excluded based on the Chalder fatigue Scale (CFS). One participant pre-CFS score is 4, which defines that he is already in a fatigue state. Other participants pre-CFS and post-CFS score was less than 4, that indicates the participants did not fatigue during the learning session. The findings of descriptive analysis, correlation analysis and statical analysis show the significance of EPM between the fatigue and non-fatigue sessions. The mental fatigue detection models show reliable and consistent performance. Notably, the Ensemble model exhibits the most consistency and reliability in the subject-dependent analysis with average accuracy 0.91, F1-score 0.91, ROC AUC score 0.96 and cross validation 0.87, where subject-independent analysis with accuracy 0.75, F1-score 0.76, ROC AUC 0.83 and cross-validation mean 0.75. This evident that the subject-dependent models are more consistent than the subject-independent model. The findings are also evident that EPM emotional states are crucial features in the mental fatigue detection for online learners. Furthermore, the findings of this research lead to proposes a real-time mental fatigue intervention framework for online learners, where mental fatigue detection is achieved using EPM emotional metrics and Ensemble learning. Future studies need to investigate the influence factors of emotional states and mental fatigue to evaluate the mental fatigue intervention framework. A comprehensive investigation is required for subject-independent and subject-dependent based mental fatigue detection and intervention techniques.