Closed sebsfox closed 3 months ago
Can accumulated local effects help with random forest directional information?
The last third of this video has an example of using ALEs, but it doesn't explain them
Decided on permutation importance. The results are similar to the model specific variable importance methods (eg the ordering of important variables is very similar), but they are consistent across different modelling methods so easier to describe to the end user
Is there a way of understanding direction of importance for RF models? Is it possible to obtain a value for importance "on the same scale" as the other model - so we have a like for like when comparing the two modelling methods?