Abstract: Selecting appropriate features to use is among the key problems in machine learning and data mining. The paper defines a new feature selection criterion， i.e. Correlation Projection Score， which can help explicitly take feature interactions into account in feature selection. Then， it presents a simple algorithm which can effectively select features according to Correlation Projection Score. Experiments show that the proposed algorithm is better than some established feature selection algorithms.

Key words: machine learning, feature selection, classification

摘要： 特征选择是机器学习中的重要研究方向。以往的特征选择方法中使用的特征或者特征集评价准则往往对属性之间的相互影响考虑较少。文章提出一种新的特征集评价准则——相关投影分（CPS），并在此基础上提出了以CPS为准则的特征选择算法。实验表明该算法取得了很好的效果。

关键词: 机器学习, 特征选择, 分类