Predicting Spatial Visualization Problems' Difficulty Level from Eye-Tracking Data.

Area of Science:

• Educational Technology
• Cognitive Science
• Human-Computer Interaction

Background:

• Adjusting learning task difficulty is crucial for effective teaching.
• E-learning platforms struggle to adapt task difficulty without teacher intervention.
• Previous research on eye-tracking for cognitive modeling used less common problem formats.

Purpose of the Study:

• To investigate differences in eye movement patterns across varying difficulty levels of spatial visualization questions.
• To develop and evaluate machine learning models for predicting question difficulty using eye-tracking data from multiple-choice questions.
• To determine if eye-tracking data is sufficient for automated difficulty assessment in e-learning.

Main Methods:

• Collected eye-tracking data from participants solving spatial visualization problems presented as multiple-choice questions.
• Applied machine learning algorithms to analyze eye movement metrics, focusing on fixation duration.
• Trained and tested predictive models on datasets with known question difficulty levels.

Main Results:

• Significant differences in eye movement were observed between questions of different difficulty levels.
• Machine learning models achieved an average accuracy of 87.60% for predicting difficulty on seen questions.
• Models demonstrated an average accuracy of 72.87% for predicting difficulty on unseen questions, indicating generalizability.

Conclusions:

• Eye movement patterns, particularly fixation duration, contain valuable information about the inherent difficulty of learning tasks.
• Machine learning models effectively utilize eye-tracking data to predict the difficulty level of spatial visualization problems.
• This approach offers a viable method for automated difficulty assessment in e-learning environments.