Pedagogical Transformation through Generative AI: A Hybrid Deep Learning Comparison with Traditional Digital Learning Materials Based on GenAI-EduNet Framework

Generative Artificial Intelligence (GenAI) tools integration in higher education have radically changed the paradigm of learning, but the empirical investigation comparing their efficiency to traditional digital goods has been severely underrepresented. GenAI-EduNet is a new hybrid deep learning architecture that, using Long Short-Term Memory (LSTM) networks, alongside multi-head self-attention transformers, predicts and analyzes learning results when students work with GenAI tools (in this case, Google NotebookLM) in comparison to traditional digital learning materials. Our study is rigorous quasi-experimental mixed-method research including 1847 undergraduate students in two academic semesters with model training and validation through Open University Learning Analytics Dataset, OULAD, and EdNet data. Some major innovations presented by our framework include: (1) a multi-modal encoding technique of engagement with gated attention which captures behavioral, cognitive and affective dimensions of learning; (2) an adaptive knowledge tracing module with transformer-based multi-head self-attention used to understand learning temporal patterns; (3) a comparative performance predictor based on dual-branch architecture with inverse propensity weighting (IPW) to perform causal inference; and (4) a binary outcome predictor to predict pass/fail. The experimental findings indicate that GenAI-EduNet classifies the outcomes of student performance with 94.7% accuracy and 0.967 AUC, which is 8.3% more than ten state-of-the-art baseline methods. The quasi-experimental analysis reveals that students using NotebookLM exhibited significantly higher learning gains across all measured outcomes: post-test scores (\(\:d=0.73\), \(\:p<.001\)), higher-order thinking skills (\(\:d=0.74\), \(\:p<.001\)), cognitive engagement (\(\:d=0.59\), \(\:p<.001\)), and self-efficacy (\(\:d=0.52\), \(\:p<.001\)). The findings of our study give practical implications related to integrating educational technology and add a proven computational model to the learning analytics research during the age of generative AI.