基于粗糙集与主成分分析的属性约简的启发式算法研究

信息系统的属性约简反映了一个决策表的本质信息,为信息系统的数据挖掘奠定的基础。生成一个决策表的所有约简或计算出最佳(定义为属性数目最少)的约简都是NP难问题。目前已有的基于粗糙集的属性约简算法由于计算复杂较高,而不适用大规模数据。本文探讨了一种基于区分矩阵和主成分分析的算法,该算法利用属性在区分矩阵中频率及属性贡献率做为启发规则,寻找最小约简,并在海关通关风险评估系统中得到良好的应用。

A Heuristic Algorithm for Attribute Reduction Based on Roush Set and PCA

Attribute reduction of information system is to remove superfluous attributes from information systems while preserving the consistency of classifications the original system provides. Identifying all reductions or the minimal reductions of an information system is already proved to be NP-hard. Therefore, heuristic rules are needed to solve this kind of NP-hard problem with higher efficiency during the reduction finding process. In this paper, we introduce some concepts of rough set relevant to reduction and present an algorithm combining discernibility matrix and principal component analysis(PCA) as heuristic knowledge to find the reduction.