Multimodal machine learning framework for predicting and enhancing higher education using reinforcement learning and graph neural networks

There has been a growing emphasis on intelligent educational systems to foster student success, with universities working to aid timely interventions and personalized learning. This study is based upon the primary research problem of developing predictions and improving student performance using new data and new approaches, especially through an integrative Adaptive Education Management Framework involving Reinforcement Learning (RL) and Graph Neural Networks (GNNs). The interesting and emergent nature of this proposed study characterized by adaptive processes through GNNs, multimodal behaviour specifications, and RL-based policies among students developing their learning approaches in real time using data. Graph Attention Networks (GATs) were used to capture changing student relationships, achieving superior performance with dropout and engagement prediction accuracy. The other component, reinforcement learning, specifically Proximal Policy Optimization (PPO), produced better dropout and engagement outcomes (39.6% drop in dropout rate, 0.37-point increase in GPAs and 22.1% increase in student engagement). Overall, the proposed system and development methods improved accuracy of dropout prediction by 17%, accurate engagement prediction by 14% and accurate identification of burnout by 22%, when compared to baselines. Overall, the integrating GNN and RL methodologies to develop adaptive, personalized education systems capable of redefining how students are supported in higher education.