基于Laplacian的稀疏化特征选择方法

Laplacian-based sparsity feature selection method

为解决传统特征选择方法在处理高维数据经常遇到“维数灾难”问题,提出了一种基于Laplacian正则化项和稀疏理论的特征选择方法(Lap-Spa-Lasso)。所提方法引入Laplacian正则化项保存样本分布信息,用于获得更具判别力特征,而方法中引入的稀疏项用于去除冗余特征。通过实验表明,所提方法不仅在三类分类器上性能表现良好,而且与现有传统特征选择方法相比,能提升分类性能并对参数表现鲁棒。

In order to solve the problem of dimensionality disaster in traditional feature selection methods,we proposed a feature selection method based on Laplacian regularization term and sparsity theory(Lap-Spa-Lasso).The proposed method uses Laplacian regularization term to preserve the distribution information of samples for obtaining more discriminant features,and the contained sparse term to remove redundant features.Experiments show that the proposed method performs well on the three classifiers,improves the classification performance,and is robust to parameters compared with the existing traditional feature selection methods.