Lets say I trained an SVM model on some training data. I then remove the support vectors from the training data and re-train the model. The new model will have new support vectors. But, what all has changed in this new model? Can I confidently say that the new model performs worse than the previous one?
[link][13 comments]