摘要
针对基于l2,p-范数的鲁棒特征选择方法存在的分类精度差问题,提出一种基于最小平方QR分解的鲁棒特征选择算法。将基于l2,p-范数的鲁棒特征选择问题转化为迭代重加权最小二乘问题,对目标函数进行求导得到权重,利用权重构造加权数据矩阵和加权类标签矩阵,最后用最小平方QR分解算法求解由两个加权矩阵构成的线性方程组问题。实验结果表明,该算法不仅收敛速度快,而且分类精度有所提高。
In view of the problem of poor classification accuracy in the robust feature selection method based on l2,p-norm,a novel algorithm based on least square QR decomposition is proposed.The robust feature selection problem based on l2,p-norm is converted into a Iterative Re-Weighted Least Squares(IRLS)problem to be solved.First,the objective function is derived to obtain the weight.Then the weighted data matrix and weighted class label matrix are constructed by the weight.Finally,the Least Square QR-factorization(LSQR)algorithm is used to solve the linear equations represented by the two weighted matrices.The experimental results show that the algorithm not only converges quickly,but also improves the classification accuracy.
作者
支晓斌
武少茹
ZHI Xiaobin;WU Shaoru(School of Science, Xi'an University of Posts and Telecommunications, Xi'an 710121, China;School of Communication and Information Engineering, Xi'an University of Posts and Telecommunications, Xi'an 710121,China)
出处
《西安邮电大学学报》
2019年第6期35-41,共7页
Journal of Xi’an University of Posts and Telecommunications
基金
国家自然科学基金资助项目(61671377,61102095,61571361,11401045)
陕西省教育厅专项科学研究计划资助项目(18JK0719)
西安邮电大学新星团队资助项目(xyt2016-01)。
