基于EDA的加权KNN分类算法  

Weighted KNN algorithm based on EDA

作　　者：谢雨寒 潘峰 Xie Yuhan;Pan Feng(School of Data Science and Information Engineering,Guizhou Minzu University,Guiyang,Guizhou 550025,China;Key Laboratory of Pattern Recognition and Intelligent System,Guizhou Minzu University)

机构地区：[1]贵州民族大学数据科学与信息工程学院,贵州贵阳550025 [2]贵州民族大学模式识别与智能系统重点实验室

出　　处：《计算机时代》2023年第8期37-40,共4页Computer Era

基　　金：贵州省教育厅教改项目(20161113006,2020047);贵州省教育厅自然科学研究项目(黔教技[2022]015号)。

摘　　要：针对传统K近邻(KNN)算法对不平衡数据集分类的不足,提出一种基于分布估计算法改进的加权KNN算法EDA-KNN。在没有先验知识的前提下,为了求解最优加权KNN算法的权重向量,构建矩阵结构种群。运用分布估计算法建立概率模型,进行采样、寻优等一系列操作,经过若干次迭代,最终获得使样本分类准确率达到最高的权重向量。通过对多个数据集进行分类,结果表明,EDA-KNN算法能够显著提升对于不平衡数据集分类的准确率,分类器性能稳定。Aiming at the deficiency of traditional KNN algorithm for classification of unbalanced data sets,an improved weighted KNN algorithm based on estimation of distribution algorithm(EDA-KNN)is proposed.A matrix structure population is constructed for solving the weight vector of the optimal weighted KNN algorithm without a priori knowledge.The probability model is established using EDA,and a series of operations such as sampling and optimization search are carried out.After several iterations,the weight vector with the highest accuracy of sample classification is finally obtained.The results of classifying multiple data sets show that EDA-KNN algorithm can significantly improve the accuracy of classification for unbalanced data sets,and the classifier performance is stable.

关 键 词：不平衡数据集 KNN算法 分布估计算法 矩阵结构 分级权重 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]