Main Article Content
Abstract
Imbalanced datasets are generated from numerous real-world applications. In datasets with imbalance the minority (positive) class is usually a small fraction of total examples, but misclassification of these examples is quite expensive. A lot of work has been done on class imbalance problem with two classes but datasets possessing multiple classes have considerably received less focus. This is because of the fact that multi-class imbalance problem is hard problem than binary class problem. Here in this work we study multi-class imbalance problem using unpruned decision trees and study the impact of resampling techniques such as random oversampling and random under sampling on the benchmark datasets. The datasets used here were imported from UCI repository and Sci-Kit, Imb-Learn machine learning tools were used for performing experiments.