Machine Learning applied to student attentiveness detection: Using emotional and non-emotional measures.

Area of Science:

• Computer Science
• Educational Technology
• Machine Learning

Background:

• E-learning lacks traditional classroom's direct student monitoring capabilities.
• Previous research explored facial features and emotions for attentiveness detection.
• A webcam-only mixed model for e-learning attentiveness was not previously tested.

Purpose of the Study:

• To develop a machine learning (ML) model for automatic student attentiveness estimation in e-learning using only a webcam.
• To aid in evaluating and improving e-learning teaching methods.
• To provide an attentiveness report for e-learning lectures.

Main Methods:

• Collected video data from seven students using personal computer webcams.
• Engineered features including eye aspect ratio (EAR), yawn aspect ratio (YAR), head pose, and emotional states (11 variables total).
• Trained and validated ML models: decision trees, random forests, SVM, and XGBoost, using human observer estimations as a benchmark.

Main Results:

• The XGBoost model demonstrated the highest performance as an attention classifier.
• Achieved an average accuracy of 80.52% and an AUROC OVR of 92.12%.
• The combination of emotional and non-emotional features yielded comparable accuracy to existing studies.

Conclusions:

• A webcam-only ML model can effectively estimate student attentiveness in e-learning.
• The developed model supports the assessment of e-learning lectures based on student engagement.
• Findings facilitate the enhancement of e-learning content and delivery through data-driven insights.