基于密度聚类和样本加权信息熵的特征选择算法

AN DENSITY CLUSTERING AND WEIGHTED INFORMATION ENTROPY BASED FEATURE SELECTION ALGORITHM

特征选择是机器学习和模式识别领域中的一个重要问题.本文提出一种非监督的特征选择算法,称为基于密度聚类和样本加权信息熵的特征选择算法(DCWIE).不同于传统的基于信息熵的特征选择算法,DCWIE使用一种加权的信息熵计算方法,增加对分类贡献大的样本的权值,并通过与聚类结合,实现无监督学习.实验结果表明了本文算法的有效性.

Feature selection （FS） is an important problem in the fields of machine learning and pattern recognition. An unsupervised feature selection algorithm is proposed in this paper, called density clustering and weighted information entropy based feature selection algorithm （DCWIE）. Unlike traditional feature selection algorithms based on information entropy, DCWIE utilizes a weighted information entropy formula to highlight those samples which are important to classification, and implements unsupervised learning by combining with clustering. Experimental results show the effectiveness of our algorithm.