一种基于改进NISD的偏二叉树马田系统的数据多分类算法  

A Data Multi-classification Algorithm of Partial Binary Tree Mahalanobis-Taguchi System Based on Improved NISD

作　　者：孙叶芳 张月义 茅婷 周慧 Sun Yefang;Zhang Yueyi;Mao Ting;Zhou Hui(School of Economics and Management,China Jiliang University,Hangzhou 310018,China)

机构地区：[1]中国计量大学经济与管理学院,杭州310018

出　　处：《统计与决策》2022年第16期22-26,共5页Statistics & Decision

基　　金：国家社会科学基金一般项目(18BJY033)。

摘　　要：为解决马田系统多分类算法存在的样本重复训练以及分类准确率下降等问题,文章提出了一种基于改进的类间相似方向数(Number of Inter-class Similarity Direction,NISD)的偏二叉树马田系统多分类算法。该算法利用马氏距离改进类间相似方向数,获得更为科学的样本分类顺序,依此顺序自上而下生成整个偏二叉树,在非叶子节点构造马田系统二分类器,生成最终的分类模型。对于含k个类别的待分类样本,该算法只用训练k-1个二分类器,便可得到马田系统多分类模型,与此同时,层层剥离样本减少了样本的重复训练。UCI数据集实验结果表明,该算法分类效率更高,分类准确率也较高。In order to solve the problems of repeated training of samples and decreasing classification accuracy in multi-classification algorithm of Mahalanobis-Taguchi system(MTS), this paper proposes a partial binary tree MTS multi-classification algorithm based on improved Number of Inter-class Similarity Direction(NISD). The algorithm uses Mahalanobis distance(MD) to improve the NISD to obtain a more scientific classification order. In this order, the entire partial binary tree structure is generated from the top to the bottom, and the MTS classifiers are arranged at each node to generate the final classification model. For the samples containing k categories, the algorithm only needs to train k-1 classifiers to obtain the MTS multi-class classification model. Meanwhile, the algorithm removes the samples layer by layer and reduces the repeated training of the samples. The experimental results of the UCI datasets show that the algorithm proposed in this paper has higher classification efficiency and accuracy.

关 键 词：马田系统 多分类 改进的类间相似方向数 偏二叉树 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]