The distance-weighted K-nearest centroid neighbor classification

作     者：Li, Ping Gou, Jianping Yang, Hebiao 

作者机构：School of Internet of Things Wuxi Institute of Technology 1600 Gaolang West Road Binhu District Wuxi214121 China School of Computer Science and Telecommunication Engineering Jiangsu University 301 Xuefu Road Jingkou District Zhenjiang212013 China 

出 版 物：《Journal of Information Hiding and Multimedia Signal Processing》 (J. Inf. Hiding Multimedia Signal Proces.)

年 卷 期：2017年第8卷第3期

页      面：611-622页

核心收录：

基　　金：This work was supported in part by National Natural Science Foundation of China (Grant No. 61502208)  the Natural Science Foundation of the Jiangsu Higher Education Institutions of China (Grant No. 14KJB520007)  China Postdoctoral Science Foundation (Grant No. 2015M570411)  Natural Science Foundation of Jiangsu Province of China (Grant No. BK20150522)  Senior Visiting Scholar Project of Jiangsu Higher Vocational College of China(Grant No. 2015FX082) and Research Foundation for Talented Scholars of JiangSu University (Grant No. 14JDG037). The authors would like to thank to the anonymous reviewers for their valuable comments and suggestions 

主　　题：Classification (of information) 

摘      要：The k-Nearest Centroid Neighbor rule is one of the effective algorithms in pattern classification. In this paper, with the goal of overcoming the sensitivity issue on the choice of the neighborhood size k and improving the classification performance, two new distance-weighted k-nearest centroid neighbor rules are proposed. According to the geometric distribution and the similarity between the nearest centroid neighbors and the query pattern, the proposed rules mainly employ the new weighted voting function to give weights in the classification voting. In order to verify the classification behavior of the proposed classifiers, we conduct extensive experiments on twelve real data sets, in comparison with the other KNN-based classifiers. Experimental results show that the new classifiers are effective algorithms for the classification tasks, owing to their satisfactory classification performance and robustness over a wide range of k. © 2017.