0
$\begingroup$

To know the importance variable in a Random Forest I used The mean decrease accuracy and mean decrease Gini.

I would like to understand what are the x-axis units of the mean decrease accuracy and mean decrease Gini on a variable importance plot obtained from a random forests classifier. For example the mean decrease accuracy and gini values range between 0,000 - 0,012 and 0-600 respectively.

enter image description here

Thanks

$\endgroup$
3
  • $\begingroup$ Please give more details, we can't answer without any context. $\endgroup$
    – Erwan
    Jun 20, 2020 at 0:13
  • $\begingroup$ I shared my plots. Mean decrease accuracy and mean decrease gini are used to know the variables importance in a RF but i don't understand the x-axis. What is the meaning of this values? How can I interpreter? $\endgroup$
    – joan llano
    Jun 20, 2020 at 0:32
  • $\begingroup$ Logically the mean decrease accuracy must be by how much the accuracy decreases when training the model without the feature. But if you want a detailed answer you should explain why you are using these measures, what is the task etc. $\endgroup$
    – Erwan
    Jun 20, 2020 at 12:44

1 Answer 1

2
$\begingroup$

Gini decrease is calculated based on the mean decrease in Gini i.e. $p_i(i-p_i)$ each time when the Tree is splitted on that Feature. Value is so high because the r package weight the impurities by the raw counts, not the proportions.

Accuracy decrease is calculated on OOB dataset by randomly shuffling the data for that particular feature in the OOB. Then subtracting the two accuracies i.e. before and after shuffling.

Unit -
Accuracy - Unitless - Ratio of correct classification to total records
Gini impurity - Unitless - Measurement of the cleanliness of split.Formal

These links have got good detail:
Stackoverflow
r-bloggers
Quora

$\endgroup$

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.