K-Segments Under Bagging approach: An experimental Study on Extremely Imbalanced Data Classification

Abstract

Imbalanced dataset could be found in many real-world domains of applications, e.g. fraud detection problem, threat detection, etc.,. There are various methods that have been proposed in the past to handle the imbalanced data classification problems, but there is still a lack of in-depth research dealing with the case of extremely imbalanced data. Moreover, we can find practically the explosion of imbalance issues within the research of big data analysis, in which the imbalance ratio increases uncontrollably. Based on these two facts, we propose in this study a simplified combination of under-sampling and ensemble learning which can adapt well with different scenarios of extreme imbalance. Experimentally, we conduct our test on 11 datasets, taken from the UCI repository and Kaggle, and can show that our proposed method is not only competitive with two common methods of Under Bagging (UB) and RUSBoost, but also very effective especially in the case of extremely imbalanced big data classification problems.