I need to know how would I get to know if I have overfitted my Machine Learning model on the train data. The performance metric I have used is Logistic Loss. Does the stability of the features affect the performance of my model? If yes, how do they relate?
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
4
You need to look for the differences in the training loss and the cross-validation and test losses. If those are low, it means the model performs fairly well. Ideally, the training loss should be roughly equal to the cross-validation and test losses. If not, the model may be overfitting.
This difference also hints at an insignificant overlap between the training data points and the cross-validation and test data points. Such features are said to be unstable. In such a case, the model only gets to see the data points in the training data and not those in the cross-validation and test data and is thereby overfitting. Hence, it performs poorly. You can verify this by computing the percentage of data points present in the cross-validation and test data from those in the training data for different features in your dataset.
answered Oct 13, 2020 at 5:59
-
1$\begingroup$ Thank you for bringing my attention to my lack of hyphenation for certain words. I appreciate your discerning eye and for helping to catch some of my grammatical mistakes/taking the time to improve my posts. I am always excited to learn more. Congrats on 1k, this site is lucky to have a dedicated editor such as yourself. Cheers. $\endgroup$– EthanCommented Sep 21, 2021 at 16:59
-
1$\begingroup$ @Ethan Thank you so much for acknowledging my minimal efforts. It is always my pleasure to be able to contribute to Stack Exchange posts so as to provide the best experience for all of its users like ourselves. More importantly, we are lucky to have this platform and major contributors like you who play their parts pretty well in improving it day by day. :) $\endgroup$ Commented Sep 22, 2021 at 7:53
-
1
-
1