Evaluating the Effectiveness of Complexity Features of Eye Movement on Computer Activities Detection

Area of Science:

• Human-Computer Interaction
• Biomedical Engineering
• Machine Learning

Background:

• Human activity recognition is crucial for health monitoring and disease prevention.
• Eye movement analysis offers a non-invasive method for understanding user engagement and cognitive states.
• Traditional eye movement analysis relies on basic statistical features, potentially missing complex behavioral patterns.

Purpose of the Study:

• To compare the efficacy of eye-movement complexity features against conventional features for detecting daily computer activities.
• To evaluate the impact of feature selection using Analysis of Variance (ANOVA) on activity recognition accuracy.
• To assess the performance of various Artificial Intelligence (AI) models in classifying computer-based tasks using eye movement data.

Main Methods:

• Collected eye movement data from 150 students performing reading, video watching, and typing tasks using a desktop eye-tracker.
• Extracted 550 complexity features (multi-scale entropy) and 56 conventional features (statistical measurements).
• Utilized ANOVA for feature screening, followed by building 12 AI models (SVM, Decision Tree, Random Forest) with feature combinations.

Main Results:

• Eye-movement complexity features achieved 85.34% accuracy, significantly outperforming conventional features (66.98%).
• ANOVA screening enhanced recognition accuracy by 2.29%.
• Random Forest models demonstrated superior performance in activity classification.

Conclusions:

• Eye-movement complexity features derived from multi-scale entropy analysis are superior for detecting computer-based human activities.
• Feature selection via ANOVA improves the accuracy of AI models for eye movement analysis.
• This research highlights the potential of advanced eye movement analysis in developing sophisticated human activity recognition systems for healthcare.