- One-vs-Rest (OvR) or One-vs-All:

    - a two class classifier trained for each class against the rest of the classes.

    -  Classifier with the highest confidence is chosen as the final prediction.

- One-vs-One (OvO):

    - binary classifiers are trained for each pair of combination of classes.

    - For n classes, n-1 classifiers are trained with pairs

    - Final output is determined by majority of votes amongst the classifiers

- Multinomial (Softmax):

    - directly trains a single classifier to predict the probabilities of each class.

    - uses a "softmax" activation function, which converts the raw outputs into class probabilities.

    - Calculates hyperplane distance with features and weights for each class. The distance is then converted to probability.

    - class with the highest probability is chosen as the final prediction.

    - nothing but the generalized argmax of multiple singular logistic regression