机器学习中的特征选择

Feature Selection in Machine Learning

20世纪90年代以来,特征选择成为机器学习领域的重要研究方向,研完成果十分显著,但是也存在许多问题需要进一步研究。本文首先对特征选择和学习算法结合的三种方式进行了系统的总结;然后将一般特征选择定位为特征集合空间中的启发式搜索问题,对特征选择算法中的四个要素进行了阐述,其中重点总结了特征评估的方法;最后对特征选择的研究现状进行了回顾,分析了目前特征选择研究的不足和未来发展的方向。

Feature selection has been an important research area in machine learning since 90's of the 20th century. Great achievements have been achieved, however many problems remain to be unsolved and need further investiga tion. In this paper,we make systematic survey on the three combination modes of featuire selection with induction al gorithm. We describe feature selection in terms of heuristic search through the space of feature sets, and discuss the foru factors in feature selection algorithms,in which the evaluation function is detailedly analyzed and discussed. Last we overview the investigation status of the feature selection ,and point out the limitations of current research and chal lenges in future work.