摘要
从属性约简后的数据集中提取规则实质上就是决策规则的约简计算,一般利用启发信息进行约简计算。提出了一种新的基于差别矩阵的决策表规则提取算法,首先从差别矩阵得到差别集,结合置信度要求得到候选规则集,然后开始提取规则并逐步调整候选规则集,最终提取出决策规则。该算法避免了规则提取过程中条件属性挑选和扩展的计算,并能够快速提取出决策表中存在的最简决策规则,计算实例表明其具有决策规则提取的工程实用性。
Extracting rules from the data set with reduced attributes is actually a reduction computation of decision rules, which is usually based on heuristic algorithm. In this paper, a new rule abstracting algorithm based on discernibility matrix has been presented. This algorithm first gets discernibility sets from discernibility matrix, generates candidate rule sets with the restriction of classification accuracy, and then begins to abstract rules from datasets. All the rules hidden in the decision dataset can be obtained efficiently by this algorithm while the computation of collecting and expanding candidate attributes doesn＇t need anymore. And case study indicates that it is valuable on the practicability of rule generation,
出处
《现代机械》
2006年第3期72-74,76,共4页
Modern Machinery
关键词
粗糙集
差别矩阵
规则提取
分类精度
rough set
discernibility matrix
rules extraction
classification accuracy
