How to see the column that influences the prediction result the most?

Question

I'm using Azure Machine Learning Studio in order to predict a column using Two-Class Boosted Decision Tree and split data.

The diagram that I have assembled can be found here:

What I need is that I'd like to see the column in the dataset that affects and influences the prediction the most. In other words, the column that changes the prediction result more than the other columns in the dataset.

Sorry if this has been asked before, but I couldn't find a proper answer to this simple question.

Answer 1

As said before, Permutation Feature Importance do the trick. Attach the Permutation Feature Importance block do the train block, click on the output port, and select visualize to get results of the module. The figure above shows the list of features sorted in descending order of their permutation importance scores.

An advice: be careful when interpreting results of permutation score when you have high correlated features.

For more info, see: https://standupdata.com/category/permutation-feature-importance/ https://gallery.cortanaintelligence.com/Experiment/Permutation-Feature-Importance-5

Answer 2

Most ML implementation for decision tree includes something called "feature importance" in its model. For example, Scikit Learn Decision Tree Classifier has an attribute that indicates the importance of each feature.

Azure ML implementation should be no exception. Please look at the below link Permutation Feature Importance.