摘要
为了提升分布不均匀样本的分类性能,该文提出了一种基于互邻信息的树型近邻(Tree-based k近邻,k Tree)分类方法,以此提高k近邻分类的准确率。首先,采用回归模型刻画样本之间的紧密程度,获取每个样本的最优k值,从而获得最优邻居,并采用k Tree提升搜索效率。其次,对于每一个测试样本,基于互邻信息准则,确定其邻域空间,完成k近邻分类。最后,数据集的试验结果表明,该文提出的基于互邻信息的k Tree分类准确率高于传统k近邻分类等其他分类算法。该文提出的方法也为k近邻分类的改进提供了新的方向。
In order to improve the classification performance of unevenly distributed samples,this paper proposes a novel tree-based k-nearest neighbor(k Tree)classification method based on mutual neighbor information,so as to improve the accuracy of k-nearest neighbor classification.Firstly,the regression model is employed to describe the closeness between samples,to obtain the optimal k value of each sample which can select optimal neighbors,and the k Tree model is applied to improve the search efficiency.Secondly,for each test sample,the neighborhood space is determined based on the mutual neighbor information criterion.And k-nearest neighbor classification is completed.Finally,the experimental results of the datasets show that the accuracy of k Tree classification based on mutual neighbor information proposed here is higher than that of other classification algorithms such as traditional k-nearest neighbor classification.The method proposed here also provides a new direction for the improvement of k-nearest neighbor classification.
作者
尹涛
胡新平
鞠恒荣
黄嘉爽
丁卫平
Yin Tao;Hu Xinping;Ju Hengrong;Huang Jiashuang;Ding Weiping(School of Information Science and Technology,Nantong University,Nantong 216002,China)
出处
《南京理工大学学报》
CAS
CSCD
北大核心
2023年第2期166-173,共8页
Journal of Nanjing University of Science and Technology
基金
国家自然科学基金(62006128,62102199,61976120)
江苏省自然科学基金(BK20191445)
江苏省双创博士计划((2020)30986)
江苏省高等学校自然科学研究重大项目(21KJA510004)
南通市科技计划项目(JC2021122)
中国博士后科学基金(2022M711716)
江苏省研究生科研与实践创新计划项目(SJCX21_1447)。
关键词
k近邻分类算法
最优邻居
回归模型
树型近邻模型
数据挖掘
k-nearest neighbor classification algorithm
optimal neighbors
regression model
k Tree model
data mining