基于最大化密度差的L_2核分类器

L_2 kernel classifier based on maximum difference of densities

提出一种基于最大化密度差的L2核分类器算法MDL2KC.该算法不仅可以保证估计出的两类密度差接近于真实密度差,而且可以使两类的密度差尽可能大.利用人工数据集和标准UCI数据集进行实验验证,所得结果表明,MDL2KC算法较传统的L2核分类器算法具有更好的分类效果和稀疏特性.

This paper proposes a kernel classification algorithm（MDL2KC） based on the theory of maximum difference of densities. MDL2KC not only ensure the estimate difference of densities fairly close to the true difference of densities, but also maximize the difference of densities between two classes. As demonstrated by extensive experiments in artifical and UCI datasets, the proposed algorithm has better classification effect and sparsity than the traditional L2 kernel classification algorithm.