Description
<--
If you want to propose a new algorithm, please refer first to the scikit-learn inclusion criterion:
https://scikit-learn.org/stable/faq.html#what-are-the-inclusion-criteria-for-new-algorithms
-->
Is your feature request related to a problem? Please describe
The time complexity of SVC on scikit-learn, I believe is O(n^2) - so therefore it takes forever to converge if a dataset is larger than 100k. The LinearSVC is more efficient to converge while it does not have _support attribute exposed, but there is a workaround I believe
Describe the solution you'd like
Allow for LinearSVC (or combined with Nystroem) to be accepted as an SVM estimator by obtaining support indices in the way described above (i.e. using decision_function)
Describe alternatives you've considered
in case data size is too big, undersample the majority class
Additional context
It has taken 10+ hours(and then timed out on VertexAI) when I tried to fit SVMSMOTE