How to determine feature importance using gradient boosting?

When using XGBoost in Python you can train a model and then use the embedded feature importance of XGBoost to determine which features are the most important.

In Matlab there is no implementation of XGBoost, but there is fitrensemble which is similar (afaik). Is there a way to use it for detemination of feature importance? Or is there maybe another way to do feature importance the way XGBoost does it?

Answer 1

[the cyclist]

The model that is output from fitrensemble has a predictorImportance method for global predictor importance.

You can also use shapley for local feature importance.

[the cyclist]

Also, note that XGBoost is not an algorithm. It's just an efficient implementation of gradient boosting. You might find this question/answer from the MathWorks support team to be interesting.