基于互信息的二进制区分矩阵特征约简方法

On Feature Reduction Algorithm of Binary Discernibility Matrix Based on Mutual Information Model

基于互信息度量的特征约简是一种重要的Filter特征约简方法,其目的是剔除条件特征中与决策类别不相关的特征,并使约简子集中特征间的相关性最小.为此,本文基于特征间的区分性评价准则,提出了互信息下二进制区分矩阵的特征约简模型,并从理论上证明了二进制区分矩阵模型与互信息模型下特征约简的等价性;其次给出了条件特征的重要性度量准则,并利用渐进式计算方法构造了一种快速的特征约简算法;最后通过实验进一步验证了算法的可行性.

Feature reduction of mutual information is one of the important approaches in Filter feature re‐duction;the purpose of mutual information is often used to delete irrelevant features and make the mini‐mum correlation in reduction subset .In this paper ,in view of discernibility between features ,the feature reduction model based on binary discernibility matrix has been presented;it is proved that feature reduc‐tion based on binary discernibility matrix is equivalent to feature reduction based on mutual information , and measure metrics of condition features has been proposed ,a quick feature reduction algorithm with In‐cremental method been designed .Finally ,the experimental results are verified to the feasibility of the algo‐rithm .