I have the following result from weka. As I observed the result I have noticed the ROC area is above 90 and the correctly classified instances is 85% Is this a sign of overfitting?
$\begingroup$
$\endgroup$
2
-
1$\begingroup$ Please avoid using screenshots in your question, it's better to copy/paste the text. $\endgroup$– ErwanJan 30, 2022 at 12:29
-
$\begingroup$ You can detect overfitting or underfitting by plotting learning curves $\endgroup$– spectreMay 15, 2022 at 5:14
Add a comment
|
2 Answers
$\begingroup$
$\endgroup$
No, this is not a sign of overfitting. The two measures (accuracy and ROC area under the curve) are not comparable.
In order to check for overfitting you need to apply the trained model on the training set itself, and then compare the performance with the one on the regular test set. There is overfitting if the training set performance is very high and the test set performance is much lower.
$\begingroup$
$\endgroup$
This seems like a good model as you are able to classify 85% of classes correctly assuming the numbers are on testing dataset.
Just by looking at one number its not possible to comment on model fit. To correctly gauge model fit please compare training set performance with test set performance and if you see a very high difference it indicates overfitting ( train perf is much better than test)
