Imbalanced Data set in machine Learning : A Comparative Study

Abstract

A system should be termed as intelligent only when it has the capability of self-learning. Machine learning being one of the most prominent field of computer science can help the system being able to get into the self-learning mode without the need of explicit programming efforts. The major challenge faced by Machine Learning experts in real life scenarios is uneven data distribution leading to imbalanced data set. Thus the proper distribution of elements in the form of sets plays a major role in achieving the self-learning goal. The uneven distribution of elements can be broadly categorized in the majority (negative) class and the minority (positive) class. The distribution of elements in nearly equal proportions is called as balanced data set. A data set is imbalanced when we have a minority class (I.e. the class which is rarer than the other classes namely the majority class). Dealing minority class is becoming more complex as classification rules tend to be fewer and weaker as compared to majority classes. Recent research findings in the area of machine learning along with the data mining have provided deeper insight into the nature of imbalanced learning along with the newer emerging challenges. Thus, this area of research is still popular among research community. In this paper we are focusing on the challenges and its best fit solutions available. Our aim to find the best fit solution by using different machine learning techniques or algorithms. These algorithms may vary in their approaches to solve the given problem. These approaches can be sampling, clustering, Graphical techniques, and statistical techniques or even with the help of classifiers. This paper provides a discussion on the complication of imbalanced data set and solutions concerning lines of future research for each of them.