非独立同分布下数值型数据的KNN算法改进  

Improved KNN algorithm for numerical data under non-independent and identical distribution

作　　者：孙聪慧 姜合[1] 相益萱 SUN Cong-hui;JIANG He;XIANG Yi-xuan(School of Computer Science and Technology,Qilu University of Technology(Shandong Academy of Sciences),Jinan 250353,China)

机构地区：[1]齐鲁工业大学(山东省科学院)计算机科学与技术学院,山东济南250353

出　　处：《计算机工程与设计》2021年第10期2816-2822,共7页Computer Engineering and Design

基　　金：国家自然科学基金青年基金项目(61502259)。

摘　　要：为挖掘数据的非独立同分布关系并解决传统KNN算法中存在的分类结果不准确的问题,提出一种非独立同分布下数值型数据的KNN改进算法。利用Pearson相关系数公式得出耦合相似度矩阵,通过该耦合相似度矩阵计算样本的类隶属度,通过ReliefF算法思想进行特征权重的计算,根据训练样本的类隶属度和特征权重更新类别决策规则,确定待分类样本的类别。对多个UCI数据集的验证结果表明,该算法能够有效提高分类准确率。To mine the non-independent and identically distributed relationship of data and solve the problem of inaccurate classification results in the traditional KNN algorithm,an improved KNN algorithm for numerical data under non-independent and identical distribution was proposed.Pearson correlation coefficient formula was used to obtain coupled similarity matrix.The samples of class membership were calculated by the similarity matrix.Through ReliefF algorithm,feature weight was calculated.According to the type of membership degree and characteristics of the training sample weights,categories of decision rules were updated,which determined the category of the classification samples.The verification results of several UCI data sets show that the proposed algorithm can effectively improve the classification accuracy.

关 键 词：非独立同分布 KNN算法 Pearson相关系数 类隶属度 特征权重 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]