基于自适应稀疏监督典型相关分析的特征选择  

FEATURE SELECTION BASED ON ADAPTIVE SPARSE SUPERVISED CANONICAL CORRELATION ANALYSIS

作　　者：王凯明[1] 鲁伊莎 肖玉柱[1] 宋学力[1] Wang Kaiming;Lu Yisha;Xiao Yuzhu;Song Xueli(School of Science,Chang’an University,Xi’an 710064,Shaanxi,China)

机构地区：[1]长安大学理学院,陕西西安710064

出　　处：《计算机应用与软件》2022年第4期294-299,共6页Computer Applications and Software

基　　金：长安大学中央高校基本科研业务费专项资金项目(310812163504)。

摘　　要：特征选择旨在识别高维数据最具信息性的特征,以实现高维数据的低维表示。稀疏监督典型相关分析模型利用样本的监督数据,通过提取具有最大相关性的稀疏典型向量实现特征选择。但是,为了求解方便,该模型一般把优化目标从典型变量的相关系数组合简化为协方差组合,此简化将导致较大的特征选择偏差。针对这一问题,提出一种新的基于自适应稀疏监督典型相关分析的特征选择模型。该模型引入一组自适应权重系数,有效解决了“两两协方差的不公平组合”问题,提高了模型的特征选择能力。实验结果验证了模型的有效性和特征选择的准确性。Feature selection is designed to identify the most informative features from high-dimensional data in order to achieve its low-dimensional representation. The sparse supervised canonical correlation analysis model utilized the supervised data to select features by searching for the sparse canonical vectors with the greatest correlation. However, for the convenience of calculation, the general optimization objective function was settled to be the combination covariance of canonical variables rather than the combination correlation coefficient, which would cause serious bias in feature selection. To solve this problem, we developed a new feature selection model based on adaptive sparse supervised canonical correlation analysis. This model introduced a set of adaptive weight coefficients to sparse supervised canonical correlation analysis model, which effectively solved the problem of “unfair combination of paired covariance” and improved the feature selection ability of the model. The experimental results verify the effectiveness of the model and the accuracy of feature selection.

关 键 词：特征选择 自适应 稀疏 监督典型相关分析 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]