How do I create a cross validated linear regression model with fitlm ?

I would like to know how I can perform cross validation on a fitlm model. All other functions for regression models support KFold as a function object. How does KFold work with fitlm as there is noch function object for cross validation implemented. I also tried crossval on a trained fitlm model which didn't work either. I hope someone can help me with my problem.

Answer 1

[Shubham Srivastava]

You can perform a K-fold cross validation for the 'fitlm' function into K folds using the 'crossval' function. In order to do so, define a predictor function handle which uses 'fitlm' and then pass the predictor function handle to the 'crossval' function.

Mentioned below is a sample code snippet to do so:

% prediction function given training and testing instances
fcn = @(Xtr, Ytr, Xte) predict( fitlm(Xtr,Ytr), Xte);
% perform cross-validation, and return average MSE across folds
mse = crossval('mse', X, Y,'Predfun',fcn, 'kfold',10);
% compute root mean squared error
avrg_rmse = sqrt(mse)

Regards,

Shubham

[Mara]

Thank you for your response! I have another question. When I have a new dataset with which I would like to predict my output using the cross validated trained model. How can I make predictions and calculet the mse then? Is it possible to create a cross validated linear regression model (as model object) like cross validated tree or SVM ?

[Mattias Blomfors]

I share the same question as Mara. How do I use the cross validated trained model to make predictions? Also, how to get the parameters for the linear model?