We know that linear kernel SVM may give better results than logistic regression since maximizing the margin usually leads to more stable results/better displacement of the decision boundary. But is there any scenario in which a linear kernel SVM performs worse than a logistic regression with respect to test accuracy?
2
-
$\begingroup$ Define "better"/"worse". By what metric? $\endgroup$– D.W.Jan 8 at 2:12
-
$\begingroup$ As written, with respect to test accuracy, meaning that lower test accuracy results in worse results $\endgroup$– DaSimJan 9 at 9:27
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
SVM may perform worse than Logistic Regression when the dataset is small, thus data points near the decision boundary (Support Vectors) may not be a true representation of the actual decision boundary, and thus may form a false maximum margin classifier boundary.
I don't have any dataset example in mind but theoretically, that should be it
