How to use various combinations of samples to build model

I am trying to use different combinations of the samples in my data(X) on each iteration to build a SVR model with fitrsvm until the model performance improves. I am not sure how to do this.

Answer 1

[Image Analyst]

The Regression Learner app on the Apps tab of the Tool ribbon will do that for you.

[NCA]

Thanks for the suggestion but the regression learner app allows for the selection of the predictor variables to include in the analysis and not the observations. I do not want to use the observations as the predictors.

I found out that I can use 'perms' but unfortuanletly the combinations of my data is too big for MATLAB to handle.

[Image Analyst]

Not true. The Regression Learner app allows you to input training data, called "predictors" and output (true answer) variables that it calls "response" but what you call "observations".

In addition you can tell it how many of the predictor rows you want to use for training, how many for validation, and how many for testing. I suggest you take another, closer, look.