基于超图稀疏的低秩属性选择算法用于多回归分析

Multi-output regression analysis algorithm based on hypergraph and low-rank

针对已有的多回归算法未能更好地考虑不同列类别之间的关联性,提出了一种基于超图稀疏的属性选择方法。利用稀疏理论中的l_(2,p)-范数来改进线性回归模型,结合低秩结构和超图表示来分别考虑同一类别间的局部结构和不同类别间的全局结构,通过子空间学习中的线性判别分析(LDA)来进一步调整重构系数矩阵的结果。经实验验证,提出的算法在六个公开数据集上相比四种对比算法,在多回归分析中均能取得更好的结果。

Since the traditional regression model does not take advantage of the correlation among class labels,this paper proposed a multi-output regression algorithm by considering hypergraph and low-rank.Firstly,this paper employed an 2,p-norm regularization term to improve the model of linear regression.Then,this paper combined low rank structure with hypergraph representation to consider both the local structure of same categories and global structure of different categories.Finally,the model further adjusted the results of reconstruction coefficient matrix by linear discriminant analysis(LDA).Experimental results show that the proposed method obtains the best results,compares to four competing methods on six public datasets.