一种基于变精度阈值的可拓熵不完备系统属性约简算法

Algorithm of Attribute Reduction Based on Extension Entropy of Variable Precision Thresholding in Incomplete Information System

针对变精度粗糙集要依据先验领域知识来确定变精度阈值,大大影响属性约简和分类效率等问题,本文将可拓学和粗糙集应用到熵理论中,并结合不完备知识系统属性约简特征,提出一种变精度阈值的可拓熵属性约简算法(RAEE),该算法充分利用可拓熵在处理动态变化数据方面优势,能动态确定因先验知识不足的变精度阈值β,并选取最优阈值区间确定属性间近似依赖为启发式标准来进行可拓熵β-近似属性约简.仿真实验结果表明,RAEE算法在不完备系统中属性约简效率较高且对噪音具有较好抗干扰性.

The problem that the variable-precision rough sets obtains from some prior domain knowledge to determine the parameter thresholding value deeply impacts on the efficiency of attribute reduction and classification.In this paper,both extension theory and rough sets are applied into the research of entropy theory,and according to characteristics of incomplete information system the algorithm（named RAEE）of attribute reduction based on extension entropy of variable precision thresholding is put forward.The algorithm can take advantage of extension entropy which can deal with dynamical data,and define variable precision thresholding β dynamically which is absent from the prior domain knowledge.Then during the scope of threshold value the algorithm can redefine attribute＇s dependence as heuristic criterion and carry on β-approximate attribute reduction based on the extension entropy.The experimental results show that algorithm of RAEE is better on the reduction efficiency and robustness at insensitivity.