Comparing the behavior of oversampling and undersampling approach of class imbalance learning by combining class imbalance problem with noise

Abstract

Class imbalance learning is a recent topic, which helps us to detect the classes from unbalanced datasets. In various real scenarios, where we need to find the exceptional cases like credit card problem, brain tumor detection, etc., the traditional classification algorithms fail because they are designed in such a way that either their results are overwhelmed by the bigger class and or they ignore the smaller class as a noise and avoid it. In recent studies, it has been found that class imbalance problem itself is not a problem but there are certain other data distribution complexities, which when combined with the class imbalance problem degrade the performance of classifier. One of the major issues is noise in the data, which is a part of every real data in one form or another. This paper compares the oversampling and undersampling approaches of class imbalance learning in noisy environment and tries to find out which is the better approach in such case.