On the Optimality of Multi-Label Classification under Subset Zero-One Loss for Distributions Satisfying the Composition Property

Maxime Gasse 1 Alex Aussem 1 Haytham Elghazel 1
1 DM2L - Data Mining and Machine Learning
LIRIS - Laboratoire d'InfoRmatique en Image et Systèmes d'information
Abstract : The benefit of exploiting label dependence in multi-label classification is known to be closely dependent on the type of loss to be minimized. In this paper, we show that the subsets of labels that appear as irreducible factors in the factor-ization of the conditional distribution of the label set given the input features play a pivotal role for multi-label classification in the context of 0/1 loss minimization, as they divide the learning task into simpler independent multi-class problems. We establish theoretical results to characterize and identify these irreducible label factors for any given probability distribution satisfying the Composition property. The analysis lays the foundation for generic multi-label classification and optimal feature subset selection procedures under this subclass of distributions. Our conclusions are supported by carefully designed experiments on synthetic and benchmark data.
Complete list of metadatas

Cited literature [27 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-01234346
Contributor : Maxime Gasse <>
Submitted on : Thursday, November 26, 2015 - 5:11:18 PM
Last modification on : Wednesday, November 20, 2019 - 2:43:06 AM
Long-term archiving on : Saturday, April 29, 2017 - 3:52:37 AM

File

gasse15.pdf
Publisher files allowed on an open archive

Identifiers

  • HAL Id : hal-01234346, version 1

Citation

Maxime Gasse, Alex Aussem, Haytham Elghazel. On the Optimality of Multi-Label Classification under Subset Zero-One Loss for Distributions Satisfying the Composition Property. International Conference on Machine Learning, Jul 2015, Lille, France. pp.2531--2539. ⟨hal-01234346⟩

Share

Metrics

Record views

515

Files downloads

319