Transformer-based deep learning for adaptive pedagogy under uncertain student preferences

Area of Science:

• Artificial Intelligence in Education
• Machine Learning for Learning Analytics
• Educational Technology

Background:

• Conventional teaching methods struggle with diverse student learning behaviors in heterogeneous educational environments.
• Individual learning preferences are often ambiguous or not explicitly expressed, posing challenges for personalization.
• There is a need for advanced AI frameworks to support adaptive learning systems.

Purpose of the Study:

• To introduce SRE-TransformerNet, an AI-driven framework for personalized and inclusive educational experiences.
• To dynamically identify and respond to variations in student learning styles.
• To enhance the robustness and effectiveness of adaptive learning systems.

Main Methods:

• Integration of Swin Transformer, ResNet, and EfficientNet within the SRE-TransformerNet framework.
• Implementation of three novel preprocessing techniques: Adaptive Range Scaling (ARS), Feature Fusion and Weight Adjustment (FFWA), and Uncertainty-Driven Transformation (UDT).
• Development of three new performance metrics: Categorical Similarity Score (CSS), Temporal Consistency Index (TCI), and Multi-Class Imbalance Metric (MCIM).

Main Results:

• SRE-TransformerNet achieved high predictive performance with an F1-score of 0.987, AUC of 0.995, and accuracy of 0.988.
• The model demonstrated a recall rate of 0.986, indicating strong efficacy in minimizing misclassifications.
• Empirical evaluations confirmed the model's robustness in handling ambiguous data and imbalanced learning scenarios.

Conclusions:

• SRE-TransformerNet effectively addresses the complexity of diverse learning styles in heterogeneous educational settings.
• The framework offers a scalable solution for enhancing educational equity and effectiveness in real time.
• SRE-TransformerNet shows strong potential for deployment in intelligent tutoring systems and large-scale e-learning platforms.