One Vs One Multi Class Classification

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

• For this one, we will train $nC2$ number of models
• Each model will be trained on pair of classes (x, y)
  • Each model will predict if the instance is of class x or y
• The final prediction will be the majority count of the prediction class

Pros:

1. Doesn't create Imbalanced dataset like One vs Rest or One vs All Multi Class Classification
2. So works better with Support Vector Machine (SVM), K-nearest Neighbor (KNN) which are dependent on data