基于决策熵的值约简算法

Value reduction algorithm based on decision information entropy

值约简是Rough集理论研究的核心内容之一,高效的值约简算法可以有助于快速做出决策.目前的值算法要么识别率不高,要么时间复杂度较高,而且也不能客观地反映决策规则的决策能力的变化情况.为了尽量克服这些缺点,文中利用置信度的概念以及决策熵能客观反映决策规则的决策能力的变化情况的优势,提出了一种基于决策熵的值约简算法.本算法采用了等价划分和容差关系在属性空间上对决策表分解,再根据置信度和决策熵判断每条决策规则中属性值是否该删除,并最终得到正确识别率上接近已有的确定规则获取算法的识别率,并且运行时间较低的结果.文中算法给出了详细地步骤以及相关实例说明,并将本算法与启发式值约简和基于决策矩阵的值约简算法做对比实验,实验结果表明,文中算法是一种可行性的值约简方法.

The research of value reduction is one of the key points in rough set theory.High efficient value reduction algorithms can help make decisions quickly.The present algorithms are either with low recognition rate or with higher time complexity,moreover,they cannot reflect the change of decision capacity of decision rules objectively.In order to overcome these shortcomings,the paper uses the advantages that the confidence level and decision information entropy can reflect the change of decision capacity of decision rules objectively,a value reduction algorithm based on decision information entropy is proposed.The algorithm uses equivalent division and tolerance relation to divide decision table in properties space,then according to confidence level and decision information entropy determine whether property value for each decision is removed or not,finally get the correct recognition rate that closes to the existing algorithm＇s,and the running time is much lower.The algorithm steps are given in detail and related examples are illustrated,and do comparative experiment with heuristic value reduction and value reduction based on decision matrix,The experiment results prove the feasibility of the proposed algorithm.