摘要
临床预测模型可用于多种医疗场景,而变量筛选是建立临床预测模型的关键步骤之一.变量筛选是从可用的候选变量列表中,筛选出对预测结局影响最大的变量,同时剔除不相关或冗余的变量.变量筛选方法大致可分为基于回归分析(向后消除法、向前筛选法、逐步筛选法、全子集筛选法、Lasso和弹性网络)和基于机器学习(随机森林、正则化随机森林、Boruta、梯度提升特征筛选)两大类.本文介绍了变量筛选的概念、流程,总结不同变量筛选方法的特点、停止规则,并比较分析各自的优缺点.
Abstract
Clinical prediction models find applications in various medical scenarios,with variable selection being one of the critical steps in their development.Variable selection involves sifting through a list of candidate variables to identify those exerting the most significant impact on predicting outcomes while discarding irrelevant or redundant ones.Variable selection method broadly fall into two categories:those based on regression analysis(backward elimination,forward selection,stepwise selection,all-possible subset selection,Lasso,and elastic net)and those rooted in machine learning(random forest,regularized random forest,Boruta,gradient boosted feature selection).This paper introduces the concept and process of variable selection,summarizes the characteristics and stop rules of different variable selection methods,and provides a comparative analysis of their respective advantages and disadvantages.
维普
万方数据