How to add feature importance in PMML

I have a data and on which I have trained decision tree. Now, these tree models give use the “feature importance” to see which features are giving the highest impact for deciding the target.

What I want is

I want to “store” those importances in the PMML? The feature importance can be stored inside MiningSchema. Will you please show me the way by which I can do that from KNIME?

Thank you.

Hi @aayushsmarten,

our XGBoost and H2O plugins have feature importance measures. For linear models (Linear Regression, Logistic Regression, GLM), the model itself can be interpreted as a feature importance.

However, the decision tree learner does not offer a feature importance.

Kind regards

1 Like

Thanks @marvin.kickuth,
I got your point. I was mistaken saying that “decision tree with importance” but my question is the same still.

How to include those in PMML?

Meaning, if I train the XGBoost model with my data and get the importance for each feature, then how can I store them in the PMML? with KNIME?

Please help, I wanted to have such information in one of my projects. It would be a huge help.

Hi @aayushsmarten,

I believe outside of the linear models it isn’t possible to store the feature importance as part of the PMML model with KNIME.

Kind regards