结合流形学习与逻辑回归的多标签特征选择  

Multi-label Feature Selection Combining Manifold Learning and Logistic Regression

作　　者：张要 马盈仓[1] 朱恒东 李恒 陈程 ZHANG Yao;MA Yingcang;ZHU Hengdong;LI Heng;CHEN Cheng(School of Science,Xi’an Polytechnic University,Xi’an 710600,China)

机构地区：[1]西安工程大学理学院,西安710600

出　　处：《计算机工程》2022年第3期90-99,106,共11页Computer Engineering

基　　金：国家自然科学基金(61976130);陕西省重点研发计划(2018KW-021);陕西省自然科学基金(2020JQ-923)。

摘　　要：对于多标签特征选择算法,通常假设数据与标签间呈现某种关系,以该关系为基础并通过正则项的约束可解决多标签特征选择问题,但该关系也可能是两种或多种关系的结合。为准确描述数据与标签间的关系并去除不相关的特征和冗余特征,基于logistic回归模型与标签流形结构提出多标签特征选择算法FSML。使用logistic回归模型的损失函数学习回归系数矩阵,利用标签流形结构学习数据特征的权重矩阵,通过L;-范数将系数矩阵和权重矩阵进行柔性结合,约束系数矩阵与权重矩阵的稀疏性并实现多标签特征选择。在经典多标签数据集上的实验结果表明,与CMLS、SCLS等特征选择算法相比,FSML算法在汉明损失、排名损失、1-错误率、覆盖率、平均精度等5个性能评价指标上表现良好,能更准确地描述数据与标签间的关系。The multi-label feature selection algorithms are usually based on an assumption that there is a certain relationship between data and labels. Based on this relationship and through the constraints of regular terms,the multilabel feature selection problem can be solved.The relationship may also be the composition of two or more relationships.To describe the relationship more accurately,a multi-tag feature selection algorithm named FSML is proposed based on the logistic regression model and label manifold structure. We use the loss function of the logistic regression model to learn the regression coefficient matrix,and use the label manifold structure to learn the weight matrix of data features.Then we employ the L;-norm to combine the coefficient matrix and weight matrix flexibly to constrain the sparsity of the matrices and realize multi-label feature selection. The experimental results on the classic multi-label datasets show that compared with feature selection algorithms such as CMLS and SCLS,the FSML algorithm performs well on five performance evaluation indicators,including Hamming Loss,Ranking Loss,One-error,Coverage,and Average Precision.It can more accurately describe the relationship between data and labels.

关 键 词：多标签学习 特征选择 LOGISTIC回归 L -范数 流形结构 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]