I am wondering whether SVM based recursive feature elimination (Guyon et al., 2000) would be considered a filter or a wrapper method.
On the one hand, it "wraps" around an SVM to get the lowest-weight features.
On the other hand, the trained model is never applied to any test set.
What do you think?
SVM-RFE is a wrapper because it features training a classifier.
Filters use some other criteria (mostly dependence tests) and the only classifier trained is the final one, built on the selected features.
- Solved – Question about Recursive feature elimination
- Solved – RFE vs Backward Elimination – is there a difference
- Solved – Feature selection : Mutual information between 2 features or between feature and target
- Solved – the difference between Recursive Feature Elimination and Backward Pass Feature Elimination
- Solved – Predictive modeling with feature selection using a small sample size