Hi Terenzio,
I understand you are using a GPR model and trying to calculate predictor importance for your predictors. Gaussian Process Regression (GPR) models do not have a built-in method for estimating predictor importance directly, however, you can use techniques like permutation importance or partial dependence plots for the same:
1.Permutation Importance: This method involves randomly shuffling one predictor variable at a time and measuring the decrease in the model's performance. The idea is that if a predictor is important, randomly shuffling its values will drastically reduce the model's performance. Here's a basic outline for the same:
- Train the GPR model on your data.
- Measure the performance of the model on a validation set
- For each predictor, create a copy of the validation set and randomly shuffle the values of the predictor. Measure the performance of the model on this shuffled validation set. The decrease in performance should be an indicator to the importance of the predictor.
2. Partial Dependence Plots (PDPs): PDPs show the dependence between the target response and a set of target features, marginalizing over the values of all other features. In essence, it can help visualize the effect of a given predictor on the output of the model, keeping all other predictors constant. You can refer to the following page for more information on how to plot partial dependence plots in MATLAB: https://www.mathworks.com/help/stats/regressiontree.plotpartialdependence.html
You can also refer to this answer to understand more on how to estimate predictor importance for a Gaussian process regression model:
I hope this helps!
