.png){kind=icon}
.png){kind=icon}
Return
SVED-SMOTE: support vector enhanced density SMOTE oversampling technique for imbalanced data classification
.png){kind=icon}
PRE
AI
J
N
DOI：10.1007/s13042-026-03240-z.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Imbalanced classification remains a significant challenge in various real-world applications, such as medical diagnosis, fraud detection, and fault diagnosis. Although the Synthetic Minority Oversampling Technique (SMOTE) has been extensively used to address imbalance, it suffers from limitations including poor representation in boundary regions and inadequate control of local data structure, resulting in reduced classifier performance. To overcome these shortcomings, this study proposes a novel oversampling method, Support Vector Enhanced Density SMOTE (SVED-SMOTE). The proposed method incorporates three key mechanisms. First, support vector information is introduced to guide the interpolation process toward discriminative boundary regions. Second, local density characteristics are utilized to adaptively adjust interpolation step sizes, enabling better representation of both dense and sparse regions. Third, a two-stage filtering strategy is employed to improve the quality of generated samples and reduce noise. Extensive experiments were conducted on 21 imbalanced datasets from the UCI repository using three evaluation metrics (G-mean, F1-score, and Area Under the ROC Curve (AUC)) with CART and SVM classifiers under five-fold cross-validation. The results demonstrate that SVED-SMOTE significantly outperforms state-of-the-art oversampling techniques across multiple metrics, particularly in datasets with complex distributions and severe class overlap. The proposed algorithm effectively improves classification performance by enhancing boundary representation and controlling noise generation, highlighting its broad applicability to diverse imbalanced learning scenarios.
Keywords:
SVED-SMOTE
Class imbalance
Oversampling
Support vectors
Local density
Classification
Journal
IF:
2.7
Papers:
3.1K
Citations:
5.6K
Organization
Cited Papers