基于属性分辨度的最大相容块规则提取算法  

Algorithm for rules acquisition from maximal consistent blocks based on attribute discernibility

作　　者：纪霞[1] 李龙澍[2] 

机构地区：[1]安徽大学计算智能与信号处理教育部重点实验室,合肥230601 [2]安徽大学计算机科学与技术学院,合肥230601

出　　处：《控制与决策》2013年第12期1837-1842,1848,共7页Control and Decision

基　　金：安徽省自然科学基金项目(60273043);安徽大学博士科研启动基金项目(33190081)

摘　　要：提出一种基于属性分辨度的不完备决策表规则提取算法,它是一种例化方向的方法.首先从空集开始,逐步选择当前最重要的条件属性对对象集分类,从广义决策值唯一的相容块提取确定规则,从其他的相容块提取不确定规则;然后设计属性必要性判断步骤去除每条规则的冗余属性;最后通过规则约简过程来简化所获得的规则,增强规则的泛化能力.实验结果表明,所提出的算法效率更高,并且所获得的规则简洁有效.An algorithm for rules acquisition from the incomplete decision table is proposed, which uses the attribute importance measure based on discernibility. This algorithm uses a method by specialization, in which condition attributes are considered to be added to selected attributes set in order of discernibility until the selected attributes set can make the classification. Certain rules are extracted from the consistent blocks with the single generalized decision, and the uncertain rules are extracted from other consistent blocks. An attribute necessity judgment step is constructed to remove redundant attributes of each rule. Besides, a rule reduction procedure is also constructed, which helps to enhance the rule generalization ability. The experiments and comparison show that the proposed algorithm can get the simple and effective rules.

关 键 词：不完备决策表 粗糙集 属性分辨度 最大相容块 规则提取 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]