Comparing Single Classifiers, Bagging, and Voting Ensembles for Predicting Students' Academic Performance

Student learning performance is a fundamental aspect of evaluating any educational system. It forms the basis for assessing quality education. This paper highlights the critical role of student academic performance for educational institutions' achievement using educational data mining (EDM). The study analyzes and assesses student performance, proposing a predictive model based on key features such as attendance and grades. It compares various classifiers, including Bayes Network, Logistic Regression, Random Forest, Support Vector Machine, and Decision Tree to categorize student performance and predict grades. Additionally, the ensemble methods Voting and Bagging were employed to enhance classifier accuracy. In evaluating the Kaggle online dataset, Voting and Bagging achieved the highest accuracy of 68%. For the local dataset, Voting, Support Vector Machine, Decision Tree, and Random Forest achieved 100% accuracy, while Bagging and Logistic Regression followed with 89% accuracy, 100% precision, and 80% recall. Naïve Bayes had the lowest performance with 56% accuracy. These results demonstrate the effectiveness of ensemble methods in educational settings and suggest their potential for further exploration. The accuracy of these techniques depends on the available data and the nature of the task. The high accuracy of the Voting classifier in predicting academic success can help educators identify at-risk students and provide necessary support, significantly improving educational outcomes. Educators and institutions can use these findings to develop targeted interventions and support systems tailored to individual student needs, ultimately promoting academic success.