Classifier Chains for Multilabel Classification

#machine learning #supervised learning #classification #multilabel

Multi-label problem

In some classification problems, we have multilabel labels to be predicted. Many different approaches are proposed to solve such problems.

Algorithm Level

Develop algorithms for multilabel problems, such as

  1. Decision trees,
  2. AdaBoost.

Problem Transformation

On problem or data level, we can transform the multi-label problem to one or more single label problems.

Binary Relevance Method

Binary relevance method, aka BM, transforms the problem into a single label problem by training a binary classifier for each label.

By doing so, the correlations between the target labels are lost.

Label Combination Method

Label combination method (label power-set method), aka CM, combines the labels into single labels.

Classifier Chain

Classifer chains, aka CC, trains $l$ classifiers where $l$ is the number of labels for each record.

Read J, Pfahringer B, Holmes G, Frank E. Classifier Chains for Multi-label Classification. 2009. pp. 254–269.

Algorithm of Classifier Chains

Read J, Pfahringer B, Holmes G, Frank E. Classifier Chains for Multi-label Classification. 2009. pp. 254–269.

Published: by ;

LM (2021). 'Classifier Chains for Multilabel Classification', Datumorphism, 03 April. Available at: https://datumorphism.leima.is/wiki/machine-learning/classification/classifier-chains/.

Current Ref:

  • wiki/machine-learning/classification/classifier-chains.md