分类判别的覆盖算法研究  

Studies on Cover Algorithm of Classification Decision

作　　者：杨金福[1] 吴福朝[1] 

机构地区：[1]中国科学院自动化研究所模式识别国家重点实验室,北京100080

出　　处：《电子与信息学报》2007年第7期1726-1730,共5页Journal of Electronics & Information Technology

基　　金：国家863计划(2003AA133060);国家重大科学工程LAMOST计划资助课题

摘　　要：基于数据的机器学习是研究从观测数据出发寻找规律,并利用这些规律对未来数据进行预测。该文提出一种新的分类判别方法——覆盖算法,其主要过程是利用某种覆盖规则算法寻找一些训练样本集的支撑点(代表点),在决策的时候仅需计算待分类样本与支撑覆盖点之间的距离并进行比较,与之最近的支撑点所在类别即为代分类样本的类别。而支撑点仅占全部训练样本的一部分,所以相比最近邻方法具有较小运算量和存储量的优点。另一方面,覆盖算法主要是样本之间的距离运算,不需要像SVM那样考虑核函数的选择问题,因此更适用于大数据量的自动分类问题。对正常星系和恒星两类光谱数据进行实验,结果表明,覆盖算法具有较好的鲁棒性、较高的分类正确率。Data-based machine learning is exploring the rule to predict new data from the observation data. In this paper, a novel classification decision method, called as Cover Algorithm （CA）, is presented. In the training procedure, some representative samples of the training set can be obtained by utilizing a certain cover rule. Then, in the classification phase, the classifier can make a decision according to the distances from a test sample to the representatives, namely the class of the test sample is determined by the representative closest to the test sample. Comparing with the nearest neighbor method, the presented method needs less cost and memory space as the representative samples are only a little part of the training set. Furthermore, cover algorithm is suitable for automated classification of large data because it does not need to consider choosing kernel function like SVM and its main computation is distance operation between samples. The experiment results show that the cover algorithm has good robustness and high classifying accuracy over Normal Galaxies and stars datasets.

关 键 词：覆盖算法 支撑覆盖 判别准则 最近邻 SVM 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]