Breiman 1984 feature importance of a variable $j$ for a regression tree $T$ is:

enter image description here

Sklearn regression tree feature importance can be get as follows: enter image description here

Where $\hat{l}_t^2$ is MSE improvement after splitting a such node $t$ and $v_t$ is the variable used to split that node. The summation means how much the variable $j$ decreased the impurity of some tree $T$. Is this equivalent (if normalized) to sklearn Feature importance ? Breiman hasn't used "Gini importance" to describe feature importance like sklearn has done, Would the expression "gini importance" be just a way to name it on normalized case?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.