On the Potential Inadequacy of Mutual Information for Feature Selection

Frénay, Benoît;Doquire, Gauthier;Verleysen, Michel
(2012) 20th International Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (ESANN 2012) — Location: Bruges (Belgium) (25.April.2012)

Files

188-Onthepotentialinadequacyofmutualinformationforfeatureselection.pdf
  • Restricted Access
  • Adobe PDF
  • 514.71 KB
Request a copy

Details

Authors
  • Frénay, BenoîtUCLouvain
    Author
  • Doquire, GauthierUCLouvain
    Author
  • Verleysen, MichelUCLouvain
    Author
Abstract
Despite its popularity as a relevance criterion for feature selection, the mutual information can sometimes be inadequate for this task. Indeed, it is commonly accepted that a set of features maximising the mutual information with the target vector leads to a lower probability of misclassification. However, this assumption is in general not true. Justifications and illustrations of this fact are given in this paper.
Affiliations

Citations

  • APA
  • Chicago
  • FWB

Frénay, B., Doquire, G., & Verleysen, M. (2012). On the Potential Inadequacy of Mutual Information for Feature Selection. Proceedings of the 20th International Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (ESANN 2012), p. 501-506. https://hdl.handle.net/2078.5/254175