一种基于多元数据图表示的特征选择优化方法

An Optimization of Feature Selection Based on Graphical Presentation of Multivariate Data

提出了一种基于多元数据图表示的特征选择优化方法。在不增加计算复杂度的基础上,类间重叠系数矩阵可以剔除方差大而分类信息差的向量,两两类间散布矩阵则用于区分类别均值和全局均值之间距离值相近的向量。实验证明该方法生成的特征向量取得的分类效果较好。

On the basis of graphical presentation of multivariate data,a feature selection optimization was presented. Sorted overlap coefficient matrix can eliminate the variable that has bigger variance and has little effect on classification,and the sorted two-two scatter matrix was used to separate the twosorted or multisorted samples when the distance of the sorted mean and the whole mean is close. Better performance was achieved when using this method to test the vegetable oil data.