Multiclass classification

This is a brief section about multiclass classification. This is the situation when we have more than two classes in our target. Models such as classification trees can handle this case using basically the same logic that we explained. For other models such as logistic regression, which is defined only for two classes, the most common approach is called One-vs-the-Rest or One-versus-All. This strategy can only be used with models that produce probabilities or other scores that can be interpreted as confidence of the classification. This method consists of fitting one classifier per class (that class versus the rest); the observations in that class will be considered the positive class and the rest the negative class. After all models have been trained, the class that is assigned to the observations is that of the model that produced the highest probability...