一种基于SVM的不平衡数据分类方法研究

Research on an unbalanced data classification method based on SVM

类分布不均衡是导致分类算法效果不佳的主要原因之一.为提升不平衡数据分类精度,提出了数据相对平衡转换与SVM集成加权相结合的分类方法.一方面考虑到类间的平衡性,将不平衡数据转化为多个相对平衡数据集;另一方面利用SVM处理二分类的优势,使用SVM作为基分类器并进行集成训练,同时利用代价敏感的训练方式平衡数据之间的类关系.选取48组KEEL官网中的不平衡数据集进行性能评估,结果表明,本方法在不平衡数据上具有良好的分类效果,为不平衡数据分类研究提供了理论参考和实用价值.

Unbalanced class distribution is one of the main reasons for the poor performance of classification algorithms.In order to improve the classification accuracy of unbalanced data,this paper gives a classification method combining data relative balance transformation and SVM integration weighting.On the one hand,considering the balance between classes,the unbalanced data is transformed into multiple relative equilibrium data sets,and another On the one hand,SVM is used to deal with the advantages of the two classifications,SVM is used as the base classifier and integrated training,and the cost-sensitive training method is used to balance the class relationship between the data.The unbalanced data sets in 48 groups of KEEL official website were selected for performance evaluation.The experimental results show that the proposed method has good classification effect on unbalanced data,and provides theoretical reference and practical value.