Predicting student's dropout in university classes using two-layer ensemble machine learning approach: A novel stacked generalization

Student dropout is a serious problem globally. It affects not only the individual who drops out but also the former school, family, and society in general. With the current development of science and technology, big data is emphasized as the most significant technology in data analysis. From the recorded educational data, efficient prediction of students' dropout is currently a hot topic of research. Previous studies have focused only on the students' dropout based on specific levels such as individual, middle school, and university level. However, ensemble learning approaches have not received much research attention so far to predict students' dropout in university classes based on rare datasets. In this paper, we propose a novel stacking ensemble based on a hybrid of Random Forest (RF), Extreme Gradient Boosting (XGBoost), Gradient Boosting (GB), and Feed-forward Neural Networks (FNN) to predict student's dropout in university classes. On the dataset collected from 2016 to 2020 at Constantine the Philosopher University in Nitra, the proposed method has demonstrated greater performance when compared with the base models using testing accuracy and the area under the curve (AUC) evaluation metrics under the same conditions. Based on the findings of this study, students at the risk of dropping out the school can be identified based on influential factors and different agents of education can refer to this information for early intervention in the uncontrolled behavior that can lead to the risk of dropping out and take proactive precautionary measures before the issue arise.