Alleviating the Class Imbalance Problem using Data Level Approach in Noisy Imbalanced Data sets

Abstract

Classification is used to detect different kinds of patterns in the data set. Although, the newlineclassification techniques are very successful in solving real-life problems but are not so newlinesuccessful with unbalanced data sets. Considering the real-time situations, sometimes it is newlinerequired to detect exceptional cases like credit card fraud, fraudulent telephone calls, shuttle newlinesystem failure, text classification, etc. If traditional classification techniques are used in newlinesuch scenarios, then it fails to detect smaller classes. Such a problem is known as a class newlineimbalance problem. newlineDue to the challenge of predicting minority class instances accurately, it is worth studying newlinehow existing models can contribute to imbalanced data sets. Class imbalance learning has newlinerecently received considerable attention in machine learning as Several existing and newlineimproved algorithms do not provide satisfactory classification performance. Standard newlinealgorithms are overwhelmed by majority examples while minority examples contribute newlinevery little. Class-imbalance learning is obligatory in many crucial areas where newlineimprovement and new ideas are always required. During the research review, it was newlineobserved that the preprocessing of data is quite important for class-imbalance problems. It newlineis important because existing classification models can be used to classify the data after newlinebalancing the data set using pre-processing approaches. Also, real-time data contains noise, newlinewhich also plays a role in deteriorating the performance of classifiers. This motivated to newlinedevelop an algorithm, that can handle the class imbalance as well as the noise problem in newlinethe data sets. The work in this thesis is to develop a method, which can work efficiently in newlinethe imbalanced environment in the presence of noise within the data. The simulations are newlinedone using MATLAB, KEEL Tool, and Python.