带有变量选择过程的分类模型误差分析

偏倚一方差分析方法是在模型选择过程中权衡模型对现有样本解释程度和未知样本估计准确度的分析方法,目的是使选定的模型检验误差尽量小.在分类或回归过程中进行有效的变量筛选可以获得更准确的模型表达,但也会因此带来一定误差.提出"选择误差"的概念,用于刻画带有变量选择的分类问题中由于变量的某种选择方法所引起的误差.将分类问题的误差分解为偏倚—方差—选择误差进行研究,考察偏倚、方差和选择误差对分类问题的总误差所产生的影响.

An Error Analysis of Classification Process with Variable Selection

Bias-Variance decomposition provides a tool to learn the explanation and generalization ability of machine learning models.Taking efficient variable selection before classification or regression process may lead to more accurate models,but also bring in new errors.The selection error mentioned here,standards for this new error,which is similar as the definition of bias and variance.Thus considering a decomposition mode of bias-variance-selection error and taking into account the influence of each one to the total error of the classification process are the main purpose of this work.