SVD-Based Mind-Wandering Prediction from Facial Videos in Online Learning.

Area of Science:

• Computer Science
• Human-Computer Interaction
• Educational Technology

Background:

• Mind-wandering in online learning reduces engagement and learning outcomes.
• Existing methods for attentional state prediction often require specialized hardware or computationally intensive gaze tracking.
• There is a need for accurate and accessible methods to monitor student attention during online learning.

Purpose of the Study:

• To develop and evaluate a novel approach for mind-wandering prediction using webcam-based eye signal analysis.
• To demonstrate the effectiveness of Singular Value Decomposition (SVD) for extracting temporal eye signals.
• To improve the accuracy of predicting attentional states in online learning environments.

Main Methods:

• Implemented a Singular Value Decomposition (SVD)-based 1D temporal eye-signal extraction method using only eye landmark detection.
• Extracted features from the SVD-processed eye signals to train a prediction model.
• Evaluated the model's performance against baseline models, including those using eye aspect ratio (EAR) and gaze tracking.

Main Results:

• The SVD-based signal effectively captures subtle and major eye movements, outperforming eye aspect ratio (EAR) signals.
• Achieved a 2% improvement in Area Under the Receiver Operating Characteristics curve (AUROC) for 'not-focus' prediction.
• Demonstrated a 7% increase in F1-score for 'not-focus' prediction compared to baseline models.

Conclusions:

• The proposed SVD-based method offers a practical and effective solution for mind-wandering prediction in webcam-based online learning.
• This approach enhances attentional state prediction accuracy without requiring specialized hardware or gaze tracking.
• The findings have significant implications for improving online educational experiences through better engagement monitoring.