基于等价支持子集重要度的质量改进算法  

Support Subset Significant Algorithm Based on Equivalence Relation for Quality Improvement

作　　者：王萌[1,2] 孙树栋[1,2] 杨宏安[1,2] 袁宗寅[1,2] 

机构地区：[1]西北工业大学系统集成与工程管理研究所,西安710072 [2]西北工业大学现代设计与集成制造技术教育部重点实验室,西安710072

出　　处：《机械工程学报》2014年第4期185-191,共7页Journal of Mechanical Engineering

基　　金：国家自然科学基金资助项目(51075337)

摘　　要：提出一种基于等价支持子集的重要度计算方法(Support subset significant based on equivalence relation,S3ER)用于计算质量特性的重要度。S3ER算法首先定义条件属性值对决策属性值的支持度,并定义条件属性值的区分能力,通过计算条件属性值区分能力的均值,得到条件属性对决策属性的重要度。S3ER算法还能够预测未知样本类别,并获得决策属性的支持子集,通过对支持子集的分析抽取决策规则。试验对比KNN算法和带有权重的KNN算法的分类精度,验证S3ER算法所得属性重要度的有效性;对比S3ER算法、带有权重的KNN算法和C4.5算法在UCI数据库上5个分类数据集的分类精度,验证S3ER算法分类的有效性。将S3ER算法应用于某航空制造企业加工数据,得出该企业的重要质量特性的属性重要度、预测样本的质量结论,并抽取质量决策规则,以改进产品质量。A support subset significant algorithm based on equivalence relation （S3ER） is presented to compute the significance of the quality characteristic. In S3ER, first, the support of conditional attribution value to the decision attribution is defined. Then, the discrimination power of the conditional attribution value is defined. Finally, the significant of conditional attribute to the decision attribute can be got by calculating the average value of the discrimination power of the conditional attribution value. At the same time, the S3ER can predict the type of data from test dataset and compute the support subset of the decision attribute. Some decision rules can be extracted from the support subset. In the experiment, the S3ER, Weighted KNN （Weighted K-nearest neighbor） and C4.5 compared their classification accuracy on five datasets from UCI repository. The S3ER is used in the manufacturing process of a Chinese aviation corporation to compute the significant, predict the test data and extract the rules for the support subset, so the quality of product of the aviation corporation can be improved.

关 键 词：质量改进 属性重要度 支持子集 等价关系 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]