电子器件2025，Vol.48Issue(4)：781-790,10.DOI:10.3969/j.issn.1005-9490.2025.04.010
基于F分布的非均匀自适应过采样方法
Non-Uniform Adaptive Oversampling Method Based on F-Distribution
摘要
Abstract
To solve the class overlapping incurred by oversampling method such as SMOTE,a non-uniform adaptive oversampling method based on F distribution,F-SMOTE,is proposed.First,proposed method calculates the density coefficient according to the distribution of majority and minority samples contained in original imbalanced dataset.Border minority samples are emphasized and the decision boundary is adjusted adaptively.Then,the oversampling mechanism is optimized based on F distribution to synthesize new sample non-uniformly.New sample appears in the safe neighborhood of original samples,avoiding the class overlapping after oversampling.Finally,the proposed method is verified using simulated dataset and 14 real scenario datasets.The experiment demonstrates that the method can improve the F1-score and AUC value of 3 classifiers and is superior to other comparative oversampling methods for more than half of datasets.关键词
过采样方法/不平衡数据/SMOTE/线性插值Key words
oversampling method/imbalanced data/SMOTE/linear interpolation分类
信息技术与安全科学引用本文复制引用
王昊宇,李鹏,郎恂..基于F分布的非均匀自适应过采样方法[J].电子器件,2025,48(4):781-790,10.基金项目
国家自然科学基金项目(62163036) （62163036）
云南省中青年学术和技术带头人后备人才项目(202105AC160094) （202105AC160094）
国家自然科学基金青年项目(62003298) （62003298）
