Do SVM suffer from imbalanced class sizes?

Question

Say I have a population with classes c1 and c2 where size(c1) >> size(c2)

Does the SVM suffer from imbalanced classes sizes where I need to manually correct the training procedure?

score 1 · Answer 1 · answered Apr 30 '17 at 01:09

1

No. The imbalanced classes are weighted inside the algorithim in such a way that if you had a label that was only 2% observations that it would be accounted for. This is why it's a common used learner for rare events.

answered Apr 30 '17 at 01:09

cutty14

76
2

score 1 · Accepted Answer · answered Apr 30 '17 at 05:43

By default, SVM is susceptible to class imbalance. In many implementations, the misclassification penalties can indeed be reweighted to account for an unequal number of samples, but this usually has to be 'turned on'. See Does Support Vector Machine handle imbalanced Dataset?, SVM for unbalanced data

Do SVM suffer from imbalanced class sizes?

2 Answers2