Imbalance Data Classification Method Based on Improved SMOTE Algorithm and Granular Computing
QiLiang Dong, Wei Lu · 2022 41st Chinese Control Conference (CCC) · 2022
Classification of imbalanced data is a problem to be solved in practical applications, and sampling algorithm is a more effective solution. The SMOTE algorithm is a classical oversampling algorithm that provides a strong guideline for dealing with imbalanced datasets, but there are still some limitations, such as blurring the boundary between positive class samples and negative class samples. To alleviate the limitations, in this study, a new oversampling method called KB-SMOTE(K-means++ Borderline-SMOTE) is proposed. The new algorithm first clusters the minority class, next synthesizes new samples between cluster center and borderline points in minority samples, which solves the problem of blurred class boundary of imbalanced datasets, avoids the interference of noisy samples and reduces the impact on the original dataset. Then, this paper uses KB-SMOTE as a data preprocessing algorithm, combined with a hypersphere information granular classifier to classify the imbalanced data. In the experimental part, the combined classifier of this paper is compared with other classification algorithms on several imbalanced datasets, and the effectiveness of the algorithm is verified.