Hello...
I have 40 samples... I want to randomly select a percentage of these 40 samples and make an equation from them by regression and test the remaining percentage of the sample with this regression equation... that is, a number of data For correlation (creation of regression equation) and some data for validation.... My question is:
1- On what basis is this sample selection percentage chosen for correlation and validation? For example, one can take 50% to 50% of the samples, one can take 70 to 30... which one is correct? Is there an article or book that tells the basics?
2- What should be the sample selection criteria for regression equation and validation?
In machine learning, a common convention is to use 70% of the observation points to constitute the training dataset, remaining 30% is split into validation set and test set.
There is no need to split data. You can use all your sample for the construction of the regression model, and then use all the sample again for internal validation with bootstrap. These are the recommendations about this topic: Internal validation of predictive models: efficiency of some procedures for logistic regression analysis - PubMed (nih.gov). These recommendations also apply for multiple linear regression, and cox regression models.