EEG complexity measures for detecting mind wandering during video-based learning

Shaohua Tang,Zheng Li

DOI: https://doi.org/10.1038/s41598-024-58889-9

IF: 4.6

2024-04-10

Scientific Reports

Abstract:This study explores the efficacy of various EEG complexity measures in detecting mind wandering during video-based learning. Employing a modified probe-caught method, we recorded EEG data from participants engaged in viewing educational videos and subsequently focused on the discrimination between mind wandering (MW) and non-MW states. We systematically investigated various EEG complexity metrics, including metrics that reflect a system's regularity like multiscale permutation entropy (MPE), and metrics that reflect a system's dimensionality like detrended fluctuation analysis (DFA). We also compare these features to traditional band power (BP) features. Data augmentation methods and feature selection were applied to optimize detection accuracy. Results show BP features excelled (mean area under the receiver operating characteristic curve (AUC) 0.646) in datasets without eye-movement artifacts, while MPE showed similar performance (mean AUC 0.639) without requiring removal of eye-movement artifacts. Combining all kinds of features improved decoding performance to 0.66 mean AUC. Our findings demonstrate the potential of these complexity metrics in EEG analysis for mind wandering detection, highlighting their practical implications in educational contexts.

multidisciplinary sciences

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to detect mind wandering (MW) through EEG complexity measures during video - based learning. Specifically, the researchers explored the effectiveness of various EEG complexity measures in detecting the state of mind wandering and compared them with the traditional band power (BP) features. The main objectives of the study include:

1. **Improve the mind wandering detection rate**: The researchers modified the traditional probe - caught method. By having the experimenter observe the participants' facial expressions and manually trigger additional probes, the capture rate of mind wandering was increased.

2. **Systematically evaluate different pre - processing pipelines and feature types**: The researchers systematically examined different EEG signal pre - processing methods and feature types. In particular, they compared the decoding performance of Multiscale Permutation Entropy (MPE) without eye - movement artifact removal with that of the band - power features after eye - movement artifact removal.

3. **Combine multiple features to improve detection accuracy**: The researchers combined 7 features (6 complexity features plus the band - power feature) and finally achieved an AUC (Area Under the Receiver Operating Characteristic Curve) detection accuracy of 0.66 through the information - theoretic feature - selection method.

Through these methods, the researchers aim to develop an EEG analysis method that can detect mind wandering in real - time in a natural environment, thereby providing support for future adaptive education systems to address the problem of distraction and improve educational effectiveness.