一种可探测新颖类别的数据流分类算法  

A data streams classification algorithm for novel class detection

作　　者：邱良佩[1] 缪裕青[1] 陈宏[1] 张锦杏[1] 

机构地区：[1]桂林电子科技大学计算机科学与工程学院,广西桂林541004

出　　处：《桂林电子科技大学学报》2013年第3期236-240,共5页Journal of Guilin University of Electronic Technology

基　　金：广西可信软件重点实验室开放基金(kx201116);广西教育厅科研项目(201204LX122)

摘　　要：针对可探测新颖类别的数据流分类算法不能处理混合属性且新颖类别探测准确率不高,引入VFDTc算法作为基分类器学习算法,并改进新颖类别探测方法以处理混合属性数据和提高新颖类别的探测准确率。实验结果表明,改进后的算法具有较高的分类模型学习效率、数据流分类精度和处理速率。Data streams classification algorithm for novel class detection can not handle the mixing properties and novel class detection accuracy is not high. In order to accelerate the classification procedure, VFDTe algorithm is used to be the base learner. A modified clustering approach is appliecl to improve the accuracy of novel class detector and extend the domains to mixed attributes. The experimental results show that the improved algorithm achieves more advanced classification accuracy and processing speed.

关 键 词：数据流 新颖类别探测 数据挖掘 分类算法 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]