一种扩充粒化的序列邻域分类方法  

An Expanded Granulation Based Sequential Neighborhood Classification Method

作　　者：亓慧[1] 杨习贝[2] 史颖 QI Hui;YANG Xibei;SHI Ying(Department of Computer,Taiyuan Normal University,Jinzhong 030619,China;School of Computer,Jiangsu University of Science and Technology,Zhenjiang 212003,China;School of Computer and Information Technology,Shanari Unizversity,Taiyuan 030006,China)

机构地区：[1]太原师范学院计算机系,山西晋中030619 [2]江苏科技大学计算机学院,江苏镇江212003 [3]山西大学计算机与信息技术学院,山西太原030006

出　　处：《山西大学学报（自然科学版）》2020年第4期885-889,共5页Journal of Shanxi University(Natural Science Edition)

基　　金：国家自然科学基金(61906078);山西省教育厅项目(J2018159,J2019163);山西省科技厅重点研发项目(201803D31055)。

摘　　要：作为邻域粒化的核心应用之一,邻域分类器因其直观的构造手段、灵活的粒度表示以及不俗的分类性能受到了众多学者的关注与推广。然而,当训练样本数目较少时,测试样本邻域粒所能提供的有限信息无法有效地支持多数投票式的标签预测。鉴于此,提出了一种扩充粒化的序列邻域分类方法。首先,设计得分评估机制对测试样本进行排序;其次,利用传统邻域分类机制对排序最为靠前的待测样本进行标注,并将其加入训练集,扩充待测样本潜在的邻域粒化空间;最终,以此种序列分类方式完成整个训练集的标签预测。对比实验结果表明,序列领域分类方法比其他基于邻域的分类方法更为有效。As one of the core applications of neighborhood granulation,neighborhood classifier was concerned and expanded due to its intuitive construction approach,flexible granularity representation and good classification performance.Nevertheless,the limited information provided by neighborhood granule of one testing sample may not support the majority voting based label prediction effectively.In view of this,an expanded granulation based sequential neighborhood classification method is proposed.The mechanism of score evaluations is designed for ranking the testing samples,and traditional neighborhood classifier is employed to label the testing sample with the highest ranking,and such sample is added into training set to expand the potential neighborhood granulation space of the following testing samples.Moreover,the predictions of the whole training set are completed by such sequential classification method.The comparative experimental results demonstrate that the proposal is better than other neighborhood granulation based classifiers.

关 键 词：粒计算 邻域分类器 邻域粒化 多数投票 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]