Improve the Effectiveness of Predicting Student Dropouts Based on Deep Learning and SMOTE Models

In the current era, the number of educational institutions is rapidly increasing, making the admissions process more challenging. Students have numerous options when choosing an educational institution to attend. Therefore, transferring schools or dropping out has become a noteworthy issue in educational institutions. Predicting student attrition and transfers holds significant importance in enhancing the quality of education and improving the teaching and learning experience. It helps educational institutions identify the core reasons leading to student attrition or transfers and provides timely and effective solutions. Predicting student performance is often a challenging task due to various factors, with one significant factor being the imbalance in real-world data collection. This article compares different data balancing techniques such as Random Over Sampling, SMOTE, and ADASYN for data preprocessing and balancing. Additionally, the article compares the effectiveness of traditional classification models like Random Forest, Support Vector Machine (SVM), K Nearest Neighbors (KNN), Naïve Bayes, Decision Tree, AdaBoost, and Gradient Boosting with newer approaches involving Deep Learning using various architectures to handle imbalanced datasets. Experimental results indicate that the combination of data balancing algorithms such as SMOTE and ADASYN with Deep Learning models yields the best predictive outcomes.