Handling Imbalanced Datasets in Researcher Scientific Classification Using Oversampling and Undersampling Approaches

The challenge of class imbalance in datasets is a prevalent issue in machine learning, often resulting in biased and inaccurate predictions. This study investigates the effectiveness of different data balancing techniques in enhancing classifier accuracy when applied to an imbalanced Indonesian National Scientific Manuscript dataset. Specifically, we examine how various classifiers respond to three widely used data balancing methods: Synthetic Minority Over-sampling Technique (SMOTE), Random Under-sampling, and Random Over-sampling. We assess the performance of six classifiers-Support Vector Machine (SVM), Logistic Regression (LR), K-Nearest Neighbors (KNN), Extreme Gradient Boosting (XGB), Naïve Bayes (NB), and Gradient Boosting (GB)-when these techniques are applied. The analysis reveals diverse outcomes, with some classifiers showing substantial accuracy gains, while others experience moderate or even adverse effects. The findings underscore the importance of selecting an appropriate balancing method based on the dataset's characteristics and the specific classification algorithm employed in the knowledge classification process.