An Improved Model Using Oversampling Technique and Cost-Sensitive Learning for Imbalanced Data Problem

Abstract

In today's world, classification learning is a vital task because of the advancement in technology. However, during the classification process, we found the classifiers (the traditional classification techniques) couldn't handle the imbalanced data, which means the instances (majority instances) that belong to one class are many more than the instances (minority instances) that belong to another class. The use of oversampling approaches and cost-sensitive strategies are two popular approaches for addressing the imbalanced class snag. However, the best outcomes are achieved by combining the two approaches. So, the paper's concentration is to propose an enhancement model by combining the cost-sensitive technique adapted from the entropy-based fuzzy support vector machine algorithm (EFSVM), called entropy-based fuzzy membership, and the oversampling method, and provide a comparison among imbalanced learning techniques on KEEL and UCI repositories. According to the experimental findings, our enhanced model will outperform all existing models in terms of performance.