Abstract: Normalized mutual information feature selection method was proposed based on quadratic Renyi's entropy. This method could efficiently estimate the mutual information, so that the computational complexity was greatly reduced. At the same time, the normalized mutual information feature selection method and embedding method were combined to get a two-stage feature selection algorithm, which could find more characteristic feature subsets. The experimental results compared with the other similar algorithms show that the proposed method can effectively reduce the computational complexity.

Key words: feature selection, mutual information, Renyi entropy, entropy estimation

摘要： 提出了一种基于二次Renyi's熵的正则化互信息特征选择方法，该方法能高效地对互信息进行估计从而使计算复杂度大大降低。同时把正则化互信息特征选择方法与嵌入式方法相结合得到一个两段式特征选择算法，该算法可以找出更具特征的特征子集。通过实验比较了该方法与其他基于互信息的特征选择算法的效率与分类精度，结果表明该方法能够有效改善计算复杂度。

关键词: 特征选择, 互信息, Renyi熵, 熵估计