How to make one_class svm working for multi-class classification?

Question

am trying to do classification using one class svm. but I want to know how can I make it working for multi-class classification?

Answer 1

SVM is basically a binary classifier, i.e build a separator, could be a line or a plane in high dimensions (very high, see kernel functions)

so to use it as a multi classes classifier you will have to train it pair-wise, meaning if you got n classes so you'll have to run it n times, each time to classify between class-n and not class-n

for multi-classification, you will have to run it n times on the data, in the best scenario n-1 runs will return not class-n and 1 will return class-n

for most cases you'll have to choose the classification based on some confidence value attached to it