覆盖决策系统的规则提取和置信度保持的属性约简算法

Algorithms of rule acquisition and confidence-preserved attribute reduction in covering decision systems

实际中收集的数据类型具有多样性,如何从这些复杂数据中获取有用的知识是人们进行数据挖掘的目标。由于覆盖粗糙集可以处理复杂的数据,基于此对覆盖决策系统的属性约简和规则提取已有不少的研究。已有的覆盖决策系统规则提取的研究只考虑唯一的置信度评估度量,然而提取的高置信度规则覆盖的样例可能较少而具有欺骗性,由此本文又引入了一个评估规则覆盖能力的度量,从而可以消除数据中的偶然因素,获取泛化能力强的高置信度规则。在此基础上,为了提取紧凑的规则,给出了一个规则置信度保持的属性约简启发式算法。

The data collected in practice is of diversity. Howto obtain useful knowledge from the complex data is the objective of data mining. Since covering rough sets can deal with complex data,there exists much study on the attribute reduction and rule acquisition of covering decision systems based on covering rough sets. The existing research on the rule acquisition of covering decision systems considered the confidence measure as the only evaluation criterion. However,the extracted high-confidence rules may cover fewer instances and then be potentially spurious. Therefore,a measure that can assess the coverage ability of rules is introduced,which can eliminate the chance in data and thus acquire high-confidence rules with more generalization ability. Furthermore,in order to extract compact rules,we propose a rule confidence-preserved attribute reduction heuristic algorithm.