Yuni Yamasari, Rafif Aydin Ahmad, Hapsari Peni Agustin Tjahyaningtijas, Anita Qoiriah, Naim Rochmawati, Agus Prihanto
The students’ academic performance is crucial since it is linked to the success of educational institutions. Consequently, a high-performance model in this area is necessary. Ensemble learning is a viable strategy for performance enhancement. This study examines the applicability of ensemble learning-stacking on the academic performance of students. Base learners are Naive Bayes and CN2 Rule Induction, whereas Logistic Regression is a meta-learner. Our research analyzes two student datasets, namely Data1 and Data2, to assess the model's dependability. In addition, the model evaluation is conducted using two methods: cross-validation and percentage split. The experimental results show that the accuracy of stacking_Data1 and stacking_Data2 dominates among the other models. Stacking_Data1 and stacking_Data2 achieve the highest accuracy levels of about 91.1% and 94.8% on the cross-validation technique and about 84.92% and 92.56% on the percentage split technique, respectively. Furthermore, these results indicate that the implementation of a stacking algorithm can improve the performance of a single learning model, as indicated by an increase in the level of accuracy of Naive Bayes_Data1, CN2 RI_Data1, Naive Bayes_Data2, and CN2 RI_Data2 in the cross-validation technique of 17.4%, 18.4%, 13.1%, and 10.5%, respectively. Using the percentage split approach, the accuracy level increases by 16.7%, 15.7%, 14.2%, and 18.9% for Naive Bayes_Data1, CN2 RI_Data1, and Naive Bayes_Data2, respectively. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.
Department of Informatics, Universitas Negeri Surabaya, Surabaya, Indonesia; Independent Software Developer, Surabaya, Indonesia; Department of Electrical Engineering, Universitas Negeri Surabaya, Surabaya, Indonesia
