Multi-Label Classification

machine-learning

Regression Methods

Classically, regression is used for binary classification given that it divides a space into two regions: true or false.

One-vs-Rest

The one-vs-rest method performs $n$ regression for $n$ possible data labels. After that, the predicted label is the label with the highest probability among all possible data labels.

Drawback: $n$ classifications must be performed for $n$ labels. Therefore, $n$ binary classifiers must be trained.

One-vs-One

Each class is compared with every other individually. It contains more details than the one-vs-rest approach due to its nature of pair-wise comparison.

Drawback: If $4$ classes are used, $6$ binary classifiers.