0
$\begingroup$

I am using the R caret package to build a randomforest classifier model for plant data.

The dataset has 7 variables - all numeric which are being used to predict if a plant will "grow" or "not grow".

This is a very simple model.

In my training dataset I have 70% of observations classified as "grow" and 30% classified at "not grow".

I have trained the model using this data and have received an accuracy of 93% and a kappa of 86%.

My question is about the Mcnemar's Test: I have a p-value of 0.8231 which I understand usually means I reject the Null hypothesis.

I have attempted to research this test and it appears to be something about proportion change before and after an event.

Would I be correct in saying this has something to do with my imbalanced proportion in my dependent variable?

Could anyone interpret this p-value?

Thank you

$\endgroup$
3
$\begingroup$

First, a p-value of 0.83 means you do not reject the null.

Second, McNemar's test is about whether the row and column marginals are equal, or, equivalently, whether the "off-diagonal" elements are equal. Since your p value is quite high, you cannot reject the null that they are equal. It's not clear, from your question, what was in the four cells of the crosstabulation.

$\endgroup$
  • $\begingroup$ Thank you for your reply. I really do appreciate it. The confusion matrix is a simple one. its a 2X2 matrix consisting of the true and false classifications for "grown" and "not grown" $\endgroup$ – Will.S89 May 20 '18 at 23:37
  • 1
    $\begingroup$ So, there are two cells that are "errors" (top right and bottom left). McNemar's tests if they are equal. Since p is high, you cannot reject the null that they are equal. $\endgroup$ – Peter Flom May 20 '18 at 23:39
  • $\begingroup$ You're welcome. The usual thing to do on this site when an answer meets your needs is to "accept" it. $\endgroup$ – Peter Flom May 21 '18 at 11:18

Your Answer

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

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