【摘要】 针对单一的不平衡数据分类算法合成样本质量不佳、未考虑类内样本分布等局限性,提出一种不平衡数据中基于权重的边界混合采样(boundary mixed sampling based on weight selection in imbalanced data,BWBMS)。剔除噪声样本并引入边界因子概念,把原样本空间分成边界集和非边界集;考虑类内样本分布,对于边界集中每个少数类样本赋予支持度权重和密度权重并增加采样比重将其划分为两类,对两类样本子集采用不同的过采样算法和过采样倍率;考虑不同区域样本重要性的不同,根据多数类样本距离其最近的k个异类近邻的平均距离来删减部分非边界集多数类样本点。实验结果表明,结合SVM分类器的BWBMS算法在不同数据集上的性能指标得到了提升,验证了其有效性。更多还原

【Abstract】 A boundary mixed sampling based on weight selection in imbalanced data(BWBMS) was proposed to solve the limitations of the single unbalanced data classification algorithm,such as poor synthetic sample quality,and the in-class sample distribution was not considered.The noise samples were eliminated and the concept of boundary factor was introduced to divide the original sample space into boundary set and non-boundary set.Considering the distribution of samples within the class,each minority sample in the boundary set was divided into two classes according to the weight of support and density,and the proportion of sampling was increased.The subsets of the two kinds of samples were divided into two classes by different oversampling algorithms and oversampling rates.Considering the different importance of samples from different regions,the majority sample points of some non-boundary sets were deleted according to the average distance between the majority sample and its nearest k heterogeneous neighbors.Experimental results show that the BWBMS algorithm combined with SVM classifier improves the performance of different data sets and verifies the effectiveness of the algorithm.更多还原