Research on Student Performance Prediction Model Based on Blended Learning Data

Abstract-Student performance prediction is a highly sought-after research direction in the field of education, and accurately predicting final exam scores is crucial for improving teaching quality. In this study, blended learning data from the “Fundamentals of Programming (C)” course at Qinghai University from 2016 to 2019 was used. By combining SMOTE (Synthetic Minority Over-sampling Technique) oversampling and Bayesian optimization algorithm, four commonly used machine learning models were employed to evaluate students’ final performance based on process-oriented learning data. The experimental results showed that these four models achieved accuracy rates of 86%, 84%, 85%, and 86% in predicting students’ grade levels. Moreover, in predicting specific final exam scores, the root mean square errors of these models were 0.35, 0.28, 0.17, and 0.17, respectively. These findings indicate that these four models can provide accurate and effective references and guidance for students and teachers, further optimizing the teaching and learning process. By comprehensively analyzing students’ process-oriented learning data, we can reveal the patterns of their academic performance in the course and use them as a basis for predicting students’ grades.