Imbalanced Class handling and Classification on Educational Dataset
Irfan Pratama, Yoga Pristyanto, Putri Taqwa Prasetyaningrum · 2021
Research that has been done related to EDM using classification methods often researchers are not concerned with the existence of the class imbalance in the dataset. Imbalanced class is a condition where there is a significant difference between the number of instances of the minority class and the number of instances of the majority class. This can cause the performance of the classification algorithm to be not optimal because the majority of classifiers can work properly when the class distribution conditions are relatively balanced in the dataset. Several studies that have been conducted state that handling an imbalanced class on the dataset is a critical step to improving the performance of the classification algorithm. This study intends to show the effect of the imbalanced data problem and find out the better resampling method to be implemented into the machine learning process. The resampling method used in this study are SMOTE, Borderline SMOTE, SMOTE - Tomek. The students’ performance dataset used as the data source and classify using several classifiers namely, Logistic Regression, K-NN, CART, Random Forest, SVM, Stacking ensemble method. The SMOTE-Tomek resampling method works best with Random Forest classification that produced 85.8% accuracy on 10-fold cross-validation and 0.89 Geometric Mean which is the best scores among other models.