Optimizing multi label student performance prediction with GNN-TINet: A contextual multidimensional deep learning...

Area of Science:

• Educational Data Mining
• Machine Learning in Education

Background:

• Accurate student performance prediction is crucial for timely educational interventions.
• Existing models struggle with complex interactions in multi-label student performance contexts.
• The California Student Performance Dataset provides rich data on demographics, behaviors, and emotional health.

Purpose of the Study:

• To develop an advanced model for precise multi-label student performance forecasting.
• To overcome limitations of prior models in capturing intricate student performance interactions.
• To enhance educational data mining for targeted interventions and improved learning outcomes.

Main Methods:

• Developed the GNN-Transformer-InceptionNet (GNN-TINet) model, integrating Graph Neural Networks (GNN), Transformer, and InceptionNet architectures.
• Applied advanced preprocessing techniques: Contextual Frequency Encoding (CFI) and Contextual Adaptive Imputation (CAI).
• Utilized a dataset comprising 97,000 student performance instances.

Main Results:

• Achieved a Predictive Consistency Score (PCS) of 0.92 and 98.5% accuracy, surpassing current benchmarks.
• Identified significant correlations between GPA, homework completion, and parental involvement.
• Demonstrated the model's capability to identify at-risk students effectively.

Conclusions:

• The GNN-TINet model offers a robust solution for multi-label student performance prediction.
• Findings support the development of focused interventions to promote educational equity.
• The study provides valuable insights for educators and policymakers to enhance learning outcomes.