An Improved Hybrid Sampling Method for Classifying imbalanced Data to Predict Student Performance

Abstract

Data imbalance in Machine Learning is defined as an unbalanced distribution of classes within a dataset. This issue is most commonly encountered in classification jobs where the distribution of classes or labels in a given dataset is not uniform. The simplest way to handle this problem is to use resampling, which involves adding records to the minority class or removing them from the majority class. In this research, we experimented with two frequently used resampling techniques: oversampling (ROS) and undersampling (RUS). We also proposed a hybrid sampling algorithm that combines ROS and RUS. To investigate all strategies, we picked a public educational dataset and used three types of machine learning algorithms: Random Forest (RF), Logistic regression (LR), and AdaBoost. One of the important conclusions of this paper is that oversampling outperforms undersampling for different classifiers and achieves greater scores in many assessment criteria. The experimental results suggest that a hybrid strategy combining ROS and RF outperforms the other benchmark techniques. This technique has a high potential for enhancing the students' performance prediction models.