SMOTE Using Border-points and Expectation-maximization Algorithm for Imbalanced Data Classification

Class imbalance refers to a classification problem which occurs when the class distribution is skewed in the dataset. It is known that class imbalance degrades the predictive performance of classifiers. As a solution to this problem, we propose GBL-SMOTE, a modified method of SMOTE technique considering the clusters and their border-points. GBL-SMOTE solves the class imbalance problem in the learning stage by generating artificial observations at the cluster boundary of minority class observations. In this study, we create an artificial class imbalanced dataset to show the effect of GBL-SMOTE compared to original SMOTE and other SMOTE-variants. In addition, through experiments on the publicly available data, it was shown as an experiment that learning a support vector machine using GBL-SMOTE has better prediction performance than using other oversampling techniques.