基于相互近邻的加权k最近邻算法  

Weighted k-nearest neighbor algorithms based on mutual nearest neighbor

作　　者：李晨 李艳颖 柴政 张宝双 LI Chen;LI Yan-ying;CHAI Zheng;ZHANG Bao-shuang(School of Mathematics and Information Science, Baoji University of Arts and Sciences, Baoji 721013, Shaanxi, China)

机构地区：[1]宝鸡文理学院数学与信息科学学院,陕西宝鸡721013

出　　处：《宝鸡文理学院学报（自然科学版）》2022年第2期15-22,共8页Journal of Baoji University of Arts and Sciences(Natural Science Edition)

基　　金：陕西省教育厅专项科研计划项目(18JK0045);宝鸡市科技计划项目(2017JH2-23);2022年度宝鸡市哲学社会科学专项课题(BJSKZX-202241);科技类横向项目:基于大数据的统计采样算法研究(2021-KJHX033)。

摘　　要：目的 设计新型加权算法提高k最近邻分类器的分类精度。方法 由于待测样本的相互近邻对待测样本类标签有更大的影响,因此在k个近邻中检测与待测样本互为近邻的邻居样本,并增大其在决策过程的权重。结果 选取UCI数据库中的17个数据集评估所提算法的性能,结果表明所提算法的精度、召回率以及查准率平均值提高0.36%~2.91%。结论 增加相互近邻的权重能够有效改进k最近邻算法的分类性能。Purposes—To design new weighted algorithm for improving the classification accuracy of k-nearest neighbor classifier.Methods—Because the mutual nearest neighbors of the samples to be tested have a greater impact on the class label of the samples to be tested,the neighbor samples that are near neighbors with the samples to be tested are detected in the k-nearest neighbors,and its weight in the decision-making process is increased.Results—17 data sets in UCI database are selected to evaluate the performance of the proposed algorithm.The results show that the accuracy,recall and precision of the proposed algorithm are improved by 0.36%to 2.91%.Conclusions—Increasing of the weight of mutual nearest neighbors can effectively improve the performance of k-nearest neighbor algorithm.

关 键 词：数据分类 K最近邻算法 相互近邻 权重 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]