Abstract: This paper presents a Granular Support Vector Machine(GSVM) learning algorithm in order to improve the performance of SVM on imbalanced datasets. The GSVM divides some granules for majority data based on granular computing theory and extracts information granules. So the data becomes balanced, then GSVM finds local support vectors from those granules. SVM learns on these LSVs together with minority data. The satisfactory generalization performance can be obtained on imbalanced data.

Key words: Guanular Support Vector Machine(GSVM), imbalanced data, information granule, local support vectors

摘要： 针对支持向量机对于非平衡数据不能进行有效分类的问题，提出一种粒度支持向量机学习算法。根据粒度计算思想对多数类样本进行粒划分并从中获取信息粒，以使数据趋于平衡。通过这些信息粒来寻找局部支持向量，并在这些局部支持向量和少数类样本上进行有效学习，使SVM在非平衡数据集上获得令人满意的泛化能力。

关键词: 粒度支持向量机, 非平衡数据, 信息粒, 局部支持向量

CLC Number: 

• TP301.6