Create a Surrogate Model for predicting outputs from 7 different variables which have been varied in a latin hypercube over a number of runs

I want to create surrogate models for an input of 7 variables to predict 2 outputs. I have a dataset of 139 separate runs which have varied the 7 variables in a Latin Hypercube to cover as much of the design space as possible. I now want to be able to estimate what the 2 outputs would be for any given random combination of the 7 input variables. Does anyone have any guidance for how to do this? I'm assuming maybe something down the lines of PCA (principal component analysis) but I'm not very familiar with this and not sure what to do with the coefficients MatLab outputs using the function pca?

I've attached the input and output datasets for reference.

Answer 1

[Arjun]

Hi @Vicki Corrigan,

I understand you're interested in building surrogate models for your dataset. However, it is important to note that PCA is a dimensionality reduction technique, not a surrogate model. While PCA can be used to simplify your data before building a surrogate model, it does not serve as a surrogate model itself.

To create a surrogate model in MATLAB, you can use methods such as "fitrgp" or "fitrtree". By evaluating the k-fold loss for each model, you can compare their performance and select the best one.

To read more about "fitrgp" and "fitrtree" you can execute "doc fitrgp" and "doc fitrtree" from inside MATLAB's command window.