中文摘要：

特征选择作为一种常见的降维方法,一直以来都是机器学习和数据挖掘领域的热点话题.针对传统特征选择算法没有充分考虑特征全局冗余性,导致选择的特征子集对分类识别精度不够高的问题,提出基于复杂网络节点重要度评估和遗传算法的特征选择算法,将每个特征视为网络节点,根据互信息建立边,将特征选择问题转化为节点重要度评估问题,利用遗传算法选择最优特征子集.实验结果表明此算法能够找到较为优秀的特征子集,有效降维并提高分类精度.

英文摘要：

Feature selection as a common method of dimensionality reduction always is one of the hot topics in machine learning and data mining field. Classic algorithms don＇t consider featurest global redun- dancies fully, which may cause classification accuracy on selected feature subset to be not high enough. For the weakness, we propose a feature selection method（FSCN） based on node importance estimation in complex networks and genetic algorithm, regarding each feature as a network node, creating edges ac- cording to mutual information, then the problem of feature selection is converted to estimate the node importance in a complex network, and choosing the best feature subset by genetic algorithm. As the ex- periment results show, our algorithm could find better feature selection subset which results in the low- dimensional data and the good classification accuracy.