文献类型：期刊文献

中文题名：基于Hellinger距离矩阵的层次型多分类算法研究

英文题名：Research on Hierarchical Multi-classification Algorithm Based on Hellinger Distance Matrix

作者：任进军[1];王宁[1]

第一作者：任进军

机构：[1]甘肃中医药大学定西校区理科教学部

第一机构：甘肃中医药大学定西校区

年份：2019

卷号：40

期号：3

起止页码：255

中文期刊名：宁夏大学学报（自然科学版）

外文期刊名：Journal of Ningxia University(Natural Science Edition)

收录：CSTPCD

基金：甘肃中医药大学定西校区校级基金资助项目(2016XJZD08,2017XJYB05,2017XJYB08)

语种：中文

中文关键词：多类别分类;类不平衡;Hellinger距离;层次模型

外文关键词：multi-class classification;imbalanced class;Hellinger distance;hierarchical model

摘要：针对层次型多分类算法中二元分类器度量标准选择不合理而导致模型分类效果较差的现象,提出基于Hellinger距离矩阵的新度量方法.结果表明,该方法具有计算简单、不受数据分布不均影响等特点,可作为类别间重叠性的衡量;依据类别间重叠性越小可分性越强的节点选择方法,用该距离可构建分类性能更优的层次模型.实验结果验证了该方法的有效性和可行性.
The article raises a new measuring method based on Hellinger distance matrix regarding the phenomena that unreasonable option of metric standard for binary classifier in Hierarchical multi- classification algorithm will cause poor model classification effect. The result shows that, this method has the features of simple computation and unaffected by uneven data distribution, and can be used as a measure of overlapping among categories;Based on the node selection method that the smaller the overlap between classes is, the more separability will be, a better classification performance hierarchical model can be built with Hellinger distance. The experimental result verifies the effectiveness and feasibility of this standard.