决定系数与相关系数辅助的LASSO回归  

LASSO Regression Assisted by Coefficient of Determination and Correlation Coefficient

作　　者：傅晨 王开军[1] FU Chen;WANG Kaijun(School of Computer and Cyber Security,Digital Fujian Environment Monitoring IoT Laboratory,Fujian Normal University,Fuzhou 350117,China)

机构地区：[1]福建师范大学计算机与网络空间安全学院,数字福建环境监测物联网实验室,福建福州350117

出　　处：《福建师范大学学报（自然科学版）》2024年第2期57-63,89,共8页Journal of Fujian Normal University：Natural Science Edition

基　　金：福建省自然科学基金资助项目(2022J01656)。

摘　　要：现有LASSO回归方法尚未解决回归关系式中冗余特征和无关特征的去除问题,提出一个决定系数与相关系数辅助的LASSO回归方法。设给定响应变量Y和备选解释变量集X,首先设计结合决定系数的LASSO回归正则化路径求解方法,找出X中的主解释变量;然后,设计结合决定系数、相关系数和正则化路径的方法,在固定主解释变量条件下求解LASSO回归的正则化路径过程中,去除X中的无关变量和冗余变量。模拟数据集和真实数据集的实验结果表明,新方法解决了LASSO回归中冗余特征和无关特征的去除问题,在冗余变量和无关变量的去除效果上胜过对比方法。Existing LASSO regression methods have not fully addressed the removal of redundant and irrelevant features in regression equation.This study proposes a LASSO regression method assisted by the coefficient of determination and correlation coefficient.Given a response variable Y and a set of candidate explanatory variables X,we first design a LASSO regression regularization path solution method that incorporates the coefficient of determination to identify the main explanatory variables in X.Then,we devise a method that combines the coefficient of determination,correlation coefficient,and regularization path to remove the irrelevant variables and redundant variables in X in the process of solving the regularization path of Lasso regression under the condition of fixed main explanatory variables.Experimental results of simulated and real datasets show that the new method effectively removes redundant and irrelevant features in LASSO regression,and outperforms the comparative methods in terms of the effect of removing redundant and irrelevant variables.

关 键 词：LASSO回归 正则化路径 决定系数 相关系数 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]