Fiducial Model Selection And Its Application

With the advent of the era of big data,model selection has become one of the hot research topics in contemporary statistics,and it has different applications in many fields,such as economics,biomedical fields and real estate.However,in practical problems,in the face of massive data,statisticians need to analyze and model the data according to actual problems,so model selection becomes a key step.In recent years,data analysis has become more and more hot,and the choice of models that follow has become more and more concerned.For the model selection method,many parameters need to be estimated.If the training data is enough,the accuracy of the model can be continuously improved,but at the cost of improving the complexity of the model,it also brings a very common problem of over-fitting.Therefore,the model selection problem needs to seek the best balance between model complexity and data concentration.Now the model selection mainly considers two theoretical methods: Bayesian model selection and model selection with adding penalty factor.Whether it is from the probability of the model or the model selected by the penalty factor,some criteria will be adopted to avoid the model.Overfitting,and at the end of this,judge the merits of the method to make better predictions and reduce unnecessary prediction errors.This paper is mainly based on multiple linear regression model,using Fiducial inference method to select the model,and using Bregman divergence which includes the advantages of many different losses to analyze the model selection results,which can compare and analyze the selected models from multiple angles.At the same time,the stability of the Fiducial model selection is tested,so that the prediction can be better.