信息观下增量式属性约简方法研究  

Incremental Attribute Reduction Algorithm Under the Information View

作　　者：刘海涛[1] 续欣莹[1] 谢珺[1] 谢刚[2] 

机构地区：[1]太原理工大学信息工程学院,太原030024 [2]太原理工大学国际教育交流学院,太原030024

出　　处：《小型微型计算机系统》2016年第2期375-380,共6页Journal of Chinese Computer Systems

基　　金：人社部留学回国人员科技活动择优资助项目(2013-68)资助;山西省自然科(2014011018-2)资助;山西省回国留学人员科研项目(2013-033;2015-045)资助

摘　　要：针对实际数据大多是动态变化的,在增加样本后,原约简集可能已不再有效,需要对其动态更新.邻域决策系统中现有的增量算法都是从代数观下分析其变化情况,本文从信息观出发,详细分析了增加样本后,条件熵的变化机制,以及其对约简集的影响规律,发现只有新增样本的不一致邻域才引起条件熵的变化,相继引起了约简集的变化.提出了一种信息观下增量式属性约简算法,该算法只需针对新增样本及其不一致邻域进行约简,有效地避免了重复约简,从而快速求得更新后的约简集.从理论上分析了算法的复杂度,通过实例表明该算法是有效的,实验进一步验证了算法的有效性和高效性.Considering most realistic data is changing, how to deal with dynamic data has become the urgent problem. The original re- duction set may be invalid and need to be updated lwhen the new simple is added to the neighborhood decision system. The existed in- cremental methods has analyzed the changing rules of the neighborhood decision from algebra view. But this paper studies the changing rules of the conditional entropy and the reduction set after adding new simple from the perspective of information, furthermore, the re- search showed that only inconsistent neighborhood of the new simple can cause these changes. An incremental attribute reduction under the information view is presented with the framework of neighborhood system, which can update the original reduction set dynamically by reducing the new simple and the inconsistent neighborhood. The presented algorithm can avoid repeated reduction effectively. The time complexity of the presented algorithm is analyzed to compare with the classical algorithm. The example and experimental results shows that the algorithm is effective and efficient by comparing with the classical algorithm.

关 键 词：条件熵 邻域系统 增量式学习 属性约简 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]