Modular analysis of dataset balancing techniques for binary classification

Abstract

This paper presents a modular analysis of dataset balancing techniques for binary classification. The study's objective is to analyze the performance or check the necessity of different balancing techniques. Several balancing techniques, such as Random Oversampling, Random Undersampling, Synthetic Minority Over-sampling Technique (SMOTE), NearMiss and Tomek Links, and their combinations are evaluated using performance metrics such as accuracy, macro F1-score, and Micro F1-Score.The results reveal that Balancing techniques are not really necessary and hardly impact the performance of models especially in binary classification tasks. It is hoped that the findings will aid practitioners in making informed decisions in selecting balancing techniques for their problem domain.