【作者】 居红云；

【导师】 李朝锋；

【作者基本信息】 江南大学 ， 计算机软件与理论， 2008， 硕士

【摘要】 聚类分析方法按一定的距离或相似性测度将数据分成若干不同的组,由此发现整个数据集合的分布、结构与模式。根据学习方式的不同可将聚类算法分为两大类:有监督聚类和无监督聚类。监督方法的分类精度通常较高,但它的结果直接依赖于所选取的训练样本,为此需要大量高质量的有标记样本,这在一定程度上限制了其应用;而且确实在很多实际应用中由于缺少形成模式类过程的知识,人们往往只能用没有类别标签的样本进行工作,即非监督分类方法,但它得到的结果通常很难令人满意。针对这种情况,本文将非监督和监督分类方法结合起来,提出一种基于K-means和SVM模型结合的全自动分类方法,主要工作总结如下:(1)对无监督聚类算法进行了讨论,着重介绍了K均值(K-means)和模糊C均值(FCM)算法,并进行了实验对比。(2)对支持向量机算法进行了深入研究,结合无监督聚类和监督分类方法的优点,提出了一种K-means和SVM结合的混合模型,Iris、Wine和遥感数据实验结果验证了混合模型的有效性。(3)深入研究了K-means算法,利用粒子群和量子粒子群算法对其进行了改进, Iris、Wine、Breast cancer和遥感数据实验结果显示了改进方法的优势。(4)将改进的K均值算法与SVM算法结合,Iris、Wine和遥感数据实验结果进一步验证了混合模型的可行性。更多还原

【Abstract】 Clustering analysis method divides data set into some groups by a certain distance or similarity measure and finds out their distributed pattern. According to the different study style, clustering method is divided into supervised and unsupervised algorithm. Classification accuracy of supervised learning algorithm is usually high, but it needs many appropriate labeled training samples that are sometimes difficultly gained, which restricts its application. Moreover sometimes in factual application, because of lacking the foreknowledge of the pattern classes, people can only use unlabeled samples, namely unsupervised classified method, but it usually gains baddish classification results. According to such situation, the paper integrates unsupervised with supervised classification algorithm, and presents a fully automatically classification method of combing K-means with SVM. The main contributions of this thesis are given as follows:(1) We discuss several unsupervised clustering algorithms, and emphatically introduce K-means and FCM algorithms, and compare and analyze their classification results for Iris, Wine and Remote sensing data set.(2) We deeply study Support Vector Machine algorithm, and propose a hybrid model of combing K-means and SVM. Experimental results for Iris data, Wine data and Remote sensing data verify the validity of our hybrid method.(3) We use Particle Swarm Optimization (PSO) and Quantum-behaved Particle Swarm Optimization (QPSO) to improve K-means algorithm, and experimental results for Iris, Wine, Breast cancer and Remote sensing data show the validity of improved algorithm.(4) We combine the improved K-means and SVM to classify original data set. Experimental results for Iris, Wine and Remote sensing data suggest the advantages of the hybrid method.更多还原