PROJECT TITLE :

Comparing Different Resampling Methods in Predicting Students Performance Using Machine Learning Techniques

ABSTRACT:

Predicting students' performance is one of the most valuable and important research areas in today's society, thanks to technological advancements. In the subject of Education, Data Mining is particularly useful for analyzing student performance. Because of the imbalanced datasets in this sector, projecting students' performance has become a difficult task, and there is no way to compare different resampling strategies. Using two different datasets, this study compares various resampling strategies such as Borderline SMOTE, Random Over Sampler, SMOTE, SMOTE-ENN, SVM-SMOTE, and SMOTE-Tomek to manage the unbalanced data problem and forecast students' performance. The distinction between multiclass and binary classification, as well as the structure of the features, are also investigated. This paper employs a variety of machine learning classifiers, including Random Forest, K-Nearest-Neighbor, Artificial Neural Network, XG-boost, Support Vector Machine (Radial Basis Function), Decision Tree, Logistic Regression, and Nave Bayes, to better assess the performance of resampling methods in solving the imbalanced problem. Model validation strategies include the Random hold-out and Shuffle 5-fold cross-validation procedures. The results obtained using various assessment measures show that models with fewer classes and nominal features will perform better. In addition, classifiers do not perform well with unbalanced data, so this issue must be addressed. Using balanced datasets improves the performance of classifiers. The Friedman test, which is a statistical significance test, also confirms that the SVM-SMOTE is more efficient than the other resampling methods. Furthermore, when utilizing SVM-SMOTE as a resampling approach, the Random Forest classifier outperformed all other models.