基于特征权重与K-Medoids算法结合的非均衡数据处理方法

Unbalanced Data Processing Method Based on Combination ofFeature Weight and K-Medoids Algorithm

目前处理非均衡数据的方法多是以重采样方法来延伸的,传统的方法在解决非均衡数据分类问题时会使样本数据分类的精确度偏向于多数类样本,而且没有解决好类内不均衡的问题,未将样本数据的特征权重考虑到分类算法或者采样方法中。因此论文提出了一种基于特征权重值与K-Medoids算法相结合的欠采样方法,这种方法解决了之前提出的问题,抽样得到的数据更有利于决策处理,从而使得分类器对不平衡数据的分类性能有所提高。通过实验表明,论文提出的方法与传统的随机欠采样方法相比,在处理相同标准数据集时具有更好分类效果,显著提高了数据集中各类的分类精度。

The current methods for dealing with unbalanced data are mostly extended by resampling methods.Traditional methods will make the accuracy of sample data classification biased towards the majority of samples when solving the problem of un⁃balanced data classification and fail to solve the imbalance within the class.The problem is that the feature weight of the sample data is not considered in the classification algorithm or sampling method.Therefore,this paper combines the feature weights of sample da⁃ta with K-Medoids algorithm,and proposes an under-sampling method based on feature weights and K-Medoids algorithm.This can overcome the problems raised before,so that the classification performance of the classifier for unbalanced data is improved.Ex⁃periments show that the method proposed in the paper has a better classification effect when dealing with the same standard data set than the random under-sampling method,and significantly improves the classification accuracy of various types of data sets.