基于AIC, BIC, CV准则的模型选择

Model Selection Based on AIC, BIC, CV Criteria

摘要: 众所周知，一个好的模型不仅要具有优良的拟合度，而且还要具有简洁的形式，那么究竟怎样平衡模型的精确度与复杂度呢，这就需要进行模型选择了。而AIC, BIC及CV恰好能平衡模型这种关系，恰好解决了当下模型选择的难题。本文依据AIC, BIC及CV准则来进行模型选择，对WAGE2数据进行建模。首先对12个变量通过简单统计量及其统计作图得到数据的一些分布特征及其相关关系。接着用AIC, BIC及CV来进行统计建模，选出最优模型，并用最小二乘法求得拟合方程，然后进行经济学意义的解释。最后为了让模型的选择更具有说服力，重复1000次实验选出最优模型，与之前的模型进行比较，得到一致最优的模型。

Abstract: As we all know, a good model must not only have a good fit, but also have a concise form, so how to balance the accuracy and complexity of the model requires model selection. And AIC, BIC and CV can just balance the relationship of the model, which just solves the problem of current model selection. This article chooses models based on AIC, BIC and CV criteria, and models WAGE2 data. Firstly, some distribution characteristics and correlations of the data are obtained through simple statistics and statistical mapping for the 12 variables. Then, we use AIC, BIC and CV to conduct statistical modeling, select the optimal model, and use least squares method to find the fitting equation, and then explain the economic significance. Finally, in order to make the selection of the model more convincing, the optimal model was selected by repeating 1000 experiments, and compared with the previous model to obtain a consistent optimal model.