改进的K-近邻模式分类-应用数学专业论文.docxVIP

摘要 摘要 I I 摘 要 在模式识别、机器学习、数据挖掘等领域中，k-近邻分类由于其直观、简单等 特点有着广泛的研究和应用背景。其应用范围已包括生物信息认证、图像的分类、 人脸识别等领域。 本文是在分析k-近邻算法的基础上，对k-近邻分类算法的改进进行了研究。主 要研究工作如下： (1)提出基于类内近邻距离加权的改进伪近邻分类算法。考虑测试样本的多个 近邻，距离近的对归属该类的影响较大，因而拥有的权值较大。对每个类别，测 试样本得到在该类经过距离加权的伪近邻，再运用近邻分类算法进行分类，从而 提高分类精度。 (2)提出了基于类均值的最近邻分类算法。在分类过程中，分类精度容易受到 离群点的影响。采用基于类均值的最近邻分类，利用每类样本的类均值信息，降 低了离群点对分类精度的影响。 (3)提出了基于局部均值的近邻分类算法。在采用近邻分类的过程中，训练样本 的数量比较少从而导致分类的准确率降低。为了提高近邻分类的分类性能，利用 测试样本在每类训练样本集的k个近邻的均值信息，从而提高在小样本下的分类精 度，同时防止训练时过拟合。 关键词：模式分类，K-近邻准则， 局部均值，类均值 ABS ABSTRACT II II ABSTRACT K- nearest neighbor classification is widely used in pattern recognition , data mining ,machine learning and other fields for its intuitive and simple. Its application has been involved in bioinformatics certification, classification images, face recognition and other fields. This article is based on k-nearest neighbor analysis to improve k- nearest neighbor classification algorithms. There are three main works as follows: This paper is mainly based on three aspects about k- neighbor classification: The neighbor classification based on distance weighted intra-class neighborhood was proposed. This classification method make use of the neighboring samples of each class and the near sample have more effect on the class, thus the sample have greater weighted distance. Then this method use the neighbor classification rule to improve the classification accuracy. The nearest neighbor classification based on class mean was proposed. In the process of classification the accuracy of nearest neighbor classifications is easily influenced by outliers. The nearest neighbor classification based on class means make full use of the class means information and reduce the impact of outliers on the classification accuracy. The nearest neighbor classification based on local mean was proposed. The small number of training samples result to low classification accuracy in the process of nearest neighbor classification. In order to improve the performance of nearest neighbor cl