基于极大团的不完备系统规则获取方法

Maximal clique-based method for acquiring association rules in incomplete system

针对不完备决策系统的规则提取问题,提出一种基于极大团的不完备系统规则获取方法。引入图中极大团概念定义相容块构造范式,将其等价转换为极小析取范式后得到不完备系统全体极大相容块,收集每一相容块最全描述即可生成极大相容块最全描述系统,进而为最全描述系统中的每一对象构造决策分辨范式得到与该对象对应的全体可信关联规则。该方法具有2个特点:针对系统中每一基本信息粒自动生成基准置信参数,避免了预设固定参数而遗漏置信度小于此参数的部分有用规则;将决策分辨范式等价变换为其极小析取范式,避免了采用特定顺序选择属性而遗漏部分有用规则。将该算法应用于某保险公司私家车客户车险数据和UCI不完备数据集,实验结果与数据分析说明了该算法的分类预测性能。

Aiming at solving the problem of rule acquisition in incomplete system, an effective algorithm based on maximal clique is proposed for getting association rules. The conception of maximal clique theory in the graph is introduced to define the normal form for constructing consistent block; all maximal consistent blocks are collected by transforming the normal form, then the most complete description system of the maximal consistent block is generated ; the decision discernible normal form is constructed for each object in the most complete description system, and all credible association rules corresponding to each object is acquired eventually. This method has two characteristics： the benchmark confidence parameter is generated for each basic information granule automatically, which would avoid omitting available rules with the confidence level less than the fixed preset parameter; the decision discernible normal form is transformed to its minimal disjunctive form equivalently, which would avoid omitting available rules due to selecting the attribute in a particular order. Finally, this algorithm is applied to incomplete data set of UCI and auto insurance data, and the experiment results show the performance of this algorithm on classifying prediction.