基于聚类和类重叠分析的近邻分类①-计算机系统应用.PDF

2015 年 第 24 卷 第 9 期 计 算 机 系 统 应 用 基于聚类和类重叠分析的近邻分类① 刘杜钢 (福建师范大学 数学与计算机科学学院, 福州 350100) 摘 要: k 近邻分类(kNN)是一种简单而有效的非参数分类算法, 但存在着参数需要人工确定, 没有显式构建分类 模型造成存储空间大、分类效率低, 且易受到“维灾”效应影响等缺点. 针对这些缺点, 提出一种高效的近邻分类 新方法, 构造了两个新的近邻分类器. 新方法使用由 K 均值聚类产生的优化的簇原型集合为分类模型, 减少了存 储空间的同时提高了分类效率; 提出三种类重叠分析策略并引入模糊基准度量以减轻维灾影响. 以该分类模型 学习方法为基础, 提出一种新的kNN 分类器和组合朴素贝叶斯的新分类器, 算法涉及的参数都可以自动确定. 在 人工和现实数据集上进行的实验表明, 新分类器具有良好的分类效率和分类准确率. 关键词: 近邻分类; K 均值聚类; 簇原型; 类重叠分析; 模糊基准度量 Neighbor Classification Based on Clustering and Class Overlapping Analysis LIU Du-Gang (School of Mathematics and Computer Science, Fujian Normal University, Fuzhou 350100, China) Abstract: K-nearest neighbor classifier (kNN) is a simple and effective non-parametric classification algorithm. The major drawbacks of the kNN include parameters to be determined manually, its low efficiency in testing phase and suffered effect of “curse of dimensionality”. An efficient method is proposed that constructing a new kNN classifier and Naive Bayes combination classifier. K-means clustering is used to build an optimal set of cluster prototype, reducing storage space while improving the classification efficiency and determining automatically parameters. Using three types of overlapping analysis strategies and fuzzy norms measure are to relieve impacts of “curse of dimensionality”. Experimental results on both synthetic and real-world data sets show that the new classifier has good classification efficiency and classification accuracy. Key words: neighbor classification; K-means clustering; cluster prototype; class overlapping analysis; fuzzy norms measure 分类是一类有监督的学习任务, 其主要思想是在 没有显式构建分类模型, 导致存储空间大和分类效率 训练阶段构建分类模型, 在分类阶段根据分类模型预 低下. 此外, kNN 易受到“维灾”效应影响, 具体表现在: