In the binary case, the probabilities are calibrated using Platt scaling 9: logistic regression on the SVM’s scores, fit by an additional cross-validation on the training data.
2021-03-29 11:19:36 246KB Plattscaling
1