I am doing incremental learning face recognition with SVM. I use SGDClassifier with loss=modifiedhuber. After I trained the model, I open camera to run real time system. I use predict function to get the predicted label. It works but if there is an unknown person detected by the camera, it will anyhow give a label.
Because of it, I think that I need to set a threshold but predict function does not have a place for me to set. So I use another function named predict_proba, which will return me the probability of each label with the detected face. But the sum of the probas is 1, so sometimes if the unknown face is detected, it will show [0,0,0,1], it is false. It makes me unable to set threshold.

Any solution to solve this????????

How to set threshold for it