用于特征筛选的最近邻(KNN)法

K Nearest Neighbor (KNN) Method Used in Feature Selection

把基于分类的最近邻 (KNN)算法用于模式识别的特征筛选过程 ,并与传统的基于线性分析的模式识别特征筛选方法主成分回归 (PCA)、偏最小二乘法 (PLS)和K-W检验等做比较 ,证明KNN方法对包容型数据的特征变量筛选尤其有效。为包容型数据的特征筛选提供了一种有力的工具。

Feature selection is a key step of data processing using pattern recognition approaches. And the data processed can be roughly divided into two types: one side type and inclusion type. Because the difference of the special distribution of samples of these two types of data, the methods used to select feature in these two types of data should be different. However, some traditional methods,such as Principal Component Regression (PCA), Partial Least Square (PLS) and so on, are usually just applicable to the one side type data. Here, the K\|Nearest Neighbor (KNN) method, one of commonly used pattern recognition classification method, is introduced for the purpose of feature selection. Practice of computation indicates that this method is not only can be used to feature selection in one\|side type data, but also more suitable than many traditional methods of feature selection when data structure is inclusion type.