Enhancing Adaptive Learning With Deep Neural Networks for Personalized Education

Abstract

This study presents a comprehensive deep learning framework that significantly advances personalized learning technologies, particularly its focus on intelligent systems and human-centered computing, Many existing adaptive educational systems struggle to integrate cognitive models, curriculum structures, and temporal learning patterns. As a result, their adaptability and interpretability are often limited when applied to diverse learner profiles. Traditional approaches generally lack robustness in handling sparse, noisy interaction data and often ignore pedagogical signals critical to effective guidance. To overcome these challenges, we propose an innovative framework that seamlessly integrates structured time-series modeling, curriculum-guided transformation strategies, and meta-supervised learning into a unified predictive system. Central to our method is the EduFormer model, a Transformer-based network equipped with knowledge-aware attention and curriculum-conditioned normalization, enabling fine-grained understanding of student behavior over time. Complementing this is Curriculum-Adaptive Meta Supervision (CAMS), a dynamic strategy that modulates training with feedback-driven adaptation and progression-aware. This dual framework not only enhances predictive accuracy but also ensures interpretability aligned with educational theories, contributing to transparent and responsible AI practices in adaptive learning systems. Experimental evaluations on benchmark datasets confirm the superior performance of our approach in personalized prediction and pedagogical coherence, demonstrating its potential for real-world deployment in next-generation adaptive learning systems.