How to find optimum parameters for a NARX model

I am using system identification toolbox to make time series prediction with NARX with 1 input and 1 target data.

I want to find optimum parameters like number of neurons, lag, number of regressors for input and output for my NARX model.

Is there any structured way to select the training data? I have 69 datasets. I combine 3 of them sequentially and use this combination as training data. Which 3 of 69 datasets should be in my training data? I want to create a NARX model which can predict all these 69 datasets with good accuracy.

Thank you in advance.

Answer 1

[Ameer Hamza]

You can use the nlarx function(): https://www.mathworks.com/help/ident/ref/nlarx.html to progammatically estimate the parameters in a for-loop for all datasets.

[Murat Yetkin]

nlarx function allow me to define parameters. my question is how to decide these parameters. is there any other way of deciding what these parameters should be?