2
$\begingroup$

Importance

A random forest model outputs the following importance values. How do I interpert them for feature selection? If it's the mean decreased accuracy does that mean that by removing them from the model the accuracy should increase?

$\endgroup$

2 Answers 2

0
$\begingroup$

I'm not sure which software you're using so I don't know the details, but generally it's simple: the highest values indicate the features which contribute the most to the target.

In particular, the mean decrease in accuracy shows how much the accuracy decreases when removing this feature. Thus again a high value (e.g. emotionality in your example) indicates an important feature for predicting the target.

$\endgroup$
0
$\begingroup$

Good afternoon @FOH, making a quick summary of the selection of resources would be to remove the resources that are not useful for anything. It does not improve model accuracy. It has no relationship with other attributes are in the set without need.

Best way to interpret it would be to plot a graph for every type you got. Facilitating your analysis of the resources you are using to improve the model. Many of these features are noise, they can also be a burden for processing when we train the model taking more time. An interesting way to make a comparison is to put the whole set to train to capture time and measure its accuracy and then select the features with the best performance, pass the model again, perform the same captures and compare whether there is an improvement or not. Carrying out these observations you could see which will be the best to select.

Searching I found a very good explanation for selection using SelectFromModel from SK-Learn, where you can perform the selection and analyze.

Feature Selection Using Random forest

SelectFromModel SK-Learn

Another very good tool that can help you is yellowbrick this tool helps a lot. I'll put a link below has a very good example with Gradient Boosting Classifier

YellowBrick Model Selection

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.