A Summary Of Cross-Validation In Model Selection

In recent years, statistical learning as a new discipline, either in theory or in applications have been a huge development, there are many major breakthrough, which have been successfully applied to pattern recognition, data mining, natural language processing, speech recognition, image recognition, information retrieval and many other computer field.Selection and evaluation of the model plays a vital role in statistical learning, the model is good or bad directly affect the accuracy of the forecasts. In the model selection and evaluation, many methods have been proposed and applied to the actual, cross-validation due to its simplicity and universality is considered an effective way, especially in less data available circumstances, through effective reuse of the data, cross-validation fully show its many advantages in the aspects of the model selection. The main idea of the cross-validation is divided data into two parts, one part used for the training of the model, and the other part used for trained model and estimates the prediction error. Finally, select the model have the minimum predict as an excellent model. In addition, due to the different ways of data segmentation and the number of segmentation, cross-validation has generated many different ways. How to select the right cross-validation method for the data in hands has become the focus of the study of people.Many scholars have conducted a lot of research for issues related to cross-validation and obtained many fruitful results, but there are still many issues have not been resolved. In this paper, we will conduct a comprehensive summary of the results of previous studies. Under a unified framework, we will get a comb of the results of previous studies and analyze the results of previous studies to provide useful clues for future researchers. We will give some regularity recommendations about how to cross-validation select model based on the existing data in the statistical learning.