A Student Dropout Risk Prediction Model Based on Supervised Learning Techniques and Large Language Models (LLMs)

Abstract Early prediction of student dropout risk is an essential but challenging task in Vietnamese higher education. This study proposes a novel model combining supervised machine learning and large language models (LLMs) to predict student dropout risk. The model utilizes structured information and unstructured data to analyze influencing factors comprehensively. By converting student data into natural language and using pre-trained LLMs, the model can understand the context and complex relationships between factors, thereby improving prediction accuracy compared to traditional methods. The study's main contributions are to propose architecture integrating LLMs into the dropout risk classification problem, identify critical factors influencing the decision to drop out and discuss the potential application of the model in practice to support early intervention.