This method is used to extend a binary class classifier to classify multiple classes.

So for $K$ classes, we will train $K$ models

• 1 model for each class
• And for model $i$, it will predict if the instance is of class $i$ or not
• In other way, class $i$ is fighting with all other or rest of the classes
• Thats why it is called one-vs-rest or one-vs-all method
• The final prediction is the $argmax$ of the all pr