带组内互斥信息的判别最小二乘特征选择方法

摘要

This paper introduces the method of discriminative least square regression of intra-group mutual exclusion information. The selection of those important features is guaranteed by considering mutual exclusion information, meanwhile, combining the universality of the feature expression. The distribution of discriminative least square regression is maximized in order to enlarge the distribution of the heterogeneous data. Thus, the distance of different categories has been expanded. In addition, intra-group mutual exclusion items enable the model as much as possible to choose those features more important and different from each other so as to obtain the significant character subsets. The experiment on the data from the real world is also made to demonstrate that the new algorithm is much better than the existing discriminative least square regression on the problem of data classification.%带组内互斥信息的判别最小二乘特征选择方法，通过考虑特征互斥信息，既保证了那些重要特征的选择，又兼顾了特征表达的广泛性。判别最小二乘模型能够最大化异类数据的分布，使不同类别的距离被扩大；另外，组内特征互斥项可以使模型尽可能地选择那些既重要、又互不相似的特征，从而获得重要的特征子集。对在现实世界中采集的数据进行实验，结果表明，带组内互斥信息的判别最小二乘特征选择方法比已有的判别最小二乘特征选择方法在数据分类问题中具有更好的效果。