A SMOTE-based quadratic surface support vector machine for imbalanced classification with mislabeled information

Abstract

Recently, Synthetic Minority Over-Sampling Technique (SMOTE) has been widely used to handle the imbalanced classification. To address the issues of existing benchmark methods, we propose a novel scheme of SMOTE based on the K-means and Intuitionistic Fuzzy Set theory to assign proper weights to the existing points and generate new synthetic points from them. Besides, we introduce the state-of-the-art kernel-free fuzzy quadratic surface support vector machine (QSSVM) to do the classification. Finally, the numerical experiments on various artificial and real data sets strongly demonstrate the validity and applicability of our proposed method, especially in the presence of mislabeled information.