Implementing multi-label classification algorithms - (work in progress)

What is multi-label classification?

• To understand multi-label classification, lets first understand simple multi-class classification that is used in most of the places.

  Multi-class Classification (MCC) : A single instance of training data is assigned to a single class amongst the group of classes {Ci} where i ∈ (0, n) where n can be any number that the problem has. MCC is very common heuristic used to classify instances of data into categories based on classes. A key observation here is, each instance can be associated with atmost one class at any given instance of time 't'.

  Multi-label classification (MLC): It is a generalization of multi-class classification where a single instance can be assigned to multiple classes aka 'labels' at a given instance of time 't'

  Understanding difference between MCC and MLC visually -

  

  ---

  

Multi-label classification algorithms

Binary Relavance Classification

Classifier Chains

Classifier Chains Random Ordering

Classifier Chains Using Monte-Carlo Algorithm