Visibility graph analysis for educational data: potentials and a case study of predicting at-risk online students

Area of Science:

• Educational Data Mining
• Network Science
• Machine Learning

Background:

• Online learning generates vast amounts of temporal data.
• Analyzing student interactions is crucial for identifying learning patterns and predicting outcomes.
• Traditional methods may not fully capture the complexity of student behavior in digital environments.

Purpose of the Study:

• To introduce and evaluate visibility graph analysis for educational time series data.
• To demonstrate its efficacy in predicting at-risk students within online learning contexts.
• To provide interpretable insights into student behavior through graph-theoretical features.

Main Methods:

• Conversion of educational time series data into visibility graphs.
• Application of graph-theoretical metrics (e.g., global efficiency, assortativity, betweenness centrality).
• Utilizing gradient boosting algorithms for classification of at-risk students.

Main Results:

• Visibility graph analysis accurately predicts at-risk online students based on clickstream data.
• Achieved classification accuracy exceeding 87% using gradient boosting.
• The methodology provides interpretable insights into student behavior, outperforming some deep learning approaches.

Conclusions:

• Visibility graph analysis is a valuable supplementary tool for educational data mining.
• It offers interpretable insights and effective prediction of learning outcomes.
• Further research is needed to optimize model and graph selection for specific educational datasets.