Noise Detection in Imbalanced Classes Using Adaptive Boosting

Abstract

One of the problems frequently encountered in machine learning is the imbalanced class problem. Since the classification algorithms in the literature show bias according to the number of observations of the classes, various methods have been developed in order to solve this problem. Existing methods in literature can lead to various problems such as generation of noise while allowing the solution of this problem. In this study, synthetic minority oversampling technique (SMOTE), which is widely used in imbalanced classification problems, is discussed with Boosting. Using the proposed approach, the noise problem after SMOTE is solved with the help of Boosting procedure. This approach was implemented using C4.5 decision tree method on 5 different datasets. As a result of the application, the average performance of F1 score, AUC and G mean increased. © 2019 IEEE.