一种新的有监督特征选择方法

A novel method for supervised feature selection

针对高维数据中的特征选择问题,提出一种有监督的特征选择方法。首先基于非线性相关度量标准作为对离散型特征进行选择,先后做选相关、去冗余两种相关分析,并采用向前方式搜索,最后用邻近算法作为分类器对所选择的特征进行实验。结果表明,该方法能选出有用的特征来提高分类准确率,并降低数据的维度。

Aiming at the problem of feature selection in high-dimensional data, a supervised feature se- lection method is proposed. It uses the nonlinear related metrics as criterion of the discrete feature selection and then relevancy and redundancy removal analysis is made. By using the forward search method, we have e- valuated the selected features with adjacent algorithm as a classifier. The results show that this method can select useful feature to improve the classification accuracy, and reduce the dimension of data.