Genetic Programming-Enabled Prediction for Students Academic Performance in Blended Learning

Blended learning has become increasingly significant for college campus courses typically since the COVID-19 pandemic outbreak. Analyzing and predicting students’ learning performance is essential for providing personalized intervention and guidance. In recent years many machine learning models have been utilized for predicting students’ academic performance, however, these supervised learning prediction models experience strong limitation in their little interpretability due to black-box characteristic. To address this issue, this paper develops a genetic programing (GP) academic performance prediction model to explicitly define the quantitative relations between students’ learning activities and final academic performance in one blended computing thinking course. A comparison is also conducted with conventional machine learning models including ANN and SVR. The experiment reveal that the proposed GP prediction model is a viable tool and can provide more satisfactory prediction performance. In additional, numerical optimization analysis of GP prediction model indicate that most critical factors that affect student’ academic perform in our course are the score of laboratory subjects, following by the score of online quizzes, class participation and online video watching time. The results can contribute to learning about students learning situations and help teacher to provide precise intervention.