Now, we will move on to the classifiers that are not inherently multiclass. If we have a classifier and it is intended for binary classification, then we need to use either one-versus-one or one-versus-all schemes.
One-versus-all classification trains classifiers for each class that predicts that an observation does or doesn't belong to that class. There should be only one classifier that predicts that an observation belongs to its respective class in the end, and that will be our prediction for the label of the observation, as illustrated in the following graph:
In comparison, one-versus-one ...
