Logistic与分类树模型变量筛选的比较——基于信用卡邮寄业务响应率分析

基于信用卡邮寄业务响应率分析来讨论Logistic模型和分类树模型在变量选取上的区别,并尝试从几个不同角度去解释两类模型变量筛选差异的原因。笔者认为没有绝对占优势的方法,需要结合具体场景和模型的特点来选择合适的模型。分类树模型在训练集上容易过度拟合,对单个变量的影响很敏感,在进行危险因素分析时结果更能强调危险因素,对孤立点的识别率很高。Logistic模型容易受到解释变量依存关系的影响,加上分类变量的影响容易过多地选入变量或者因子,对孤立点敏感,对噪点不敏感。判别函数的差异是变量筛选差异的关键因素。

Comparison Analysis on Logistic Regression and Tree Models:Based on Response Ratio of Credit Mail Advertising

This article tries to give a comparison analysis on Logistic regression and Tree models based on response ratio of credit mail advertising.There is no good models for all situation.Tree models tend to be overfitting on training set.They could be used in risk factor discrimination because these models are sensible to single variable.Tree models are sensiable to outliers.Logistic regression models tend to use some unnecessary dependent variables,for these models are sensible to the dependency between different variables and factors.They are sensible to outliers but not to noise.The difference between discriminant functions is the key factor in explaining the difference between Logistic regression and Tree models.