Stacking Integration Strategy-Based Learning Method for Prediction of Performance in Exams

Academic performance serves as a direct indicator of learning effectiveness. In the paper, a stacking learning method is proposed to predict students’ performance in exam. Firstly, an in-depth correlation analysis is conducted to examine various factors, including students’ school situation and exam scores in other subjects, which have an impact on the exam scores in a specific subject. It is found that a strong correlation exists between exam scores in different subjects, and factors such as students’ lunch level and parents’ education level significantly impact these scores. Secondly, the stacking integration strategy employs Random Forest (RF) as the base learner, while the random gradient model is selected as the meta learner to construct a predictive model for learning achievement. Finally, a comparative experiment is conducted using multiple models, including Decision Trees (DT) and support vector machines (SVM), to predict students’ exam scores. The results show that the learning achievement prediction model based on the stacking integration strategy outperforms all other tested models, exhibiting an accuracy improvement ranging from 6% to 17%. Moreover, its performance in statistical evaluation indicators and error evaluation indicators are well. The proposed learning method is more efficient to predict students’ performance in exams.