一种最小测试代价约简的改进算法  

An Improved Algorithm of Minimal Test-cost Attributes Reduction in Decision Systems

作　　者：何华平[1] 陈光建[1] 

机构地区：[1]四川理工学院计算机学院,四川自贡643000

出　　处：《郑州大学学报（理学版）》2015年第1期74-77,86,共5页Journal of Zhengzhou University:Natural Science Edition

基　　金：国家自然科学基金资助项目;编号60873077;企业信息化与物联网测控技术四川省高校重点实验室项目;编号2013WZY02;四川理工学院培育项目;编号2013PY06;四川省教育厅科研项目;编号12ZB083

摘　　要：传统属性约简的目标是在决策表中的所有条件属性中,选择一组分类代价最小的约简,算法构建了测试代价最小的约简.以往的测试代价约简算法查找成功率不够理想,性能不稳定,提出了一种改进的测试代价约简算法.通过运行2个UCI数据集实验,证明算法是有效的,并为提高测试代价约简算法性能提供了新途径.The traditional attribute reduction problem was aimed at constructing a minimal reduct. As lat- er on, it aimed at constructing a minimal test-cost reduct under the model. The current test-cost reduction algorithm could not ensure a satisfactory performance. And an enhanced algorithm was designed for better performance. The experimental results in two UCI datasets showed that this algorithm was efficient and ef- fective. This algorithm was a new way to improve the performance.

关 键 词：代价敏感学习 属性约简 最小测试代价 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]