回归分析中内外符合精度指标研究

Research on the training&test data precision index in regression analysis

为了改正目前用于评定回归分析模型的内、外符合精度指标存在的缺陷,提出利用误差理论,依据回归分析中建模数据的模型残差和检验数据的模型残差,分别推导出计算观测值单位权方差真值之无偏估计值的公式,并将其分别定义为模型内、外符合精度检验的指标,且认为这两个指标具有可比性,可以代替当前所用的内、外符合精度指标推广使用。

In order to overcome the defects of the current training and test data precision index used to estimate the precision of regression analysis model,this paper applies the theory of errors to derived respectively the formulas to estimate the unbiased estimators of the true value of the unit weight variance of the observed values,based on the model residuals of the training data and the ones of the test data.The two unbiased estimators are defined as training and test data precision index of the regression analysis function model.This paper presents that the two indexes have comparability,which can be used to replace the current training and test data precision index.