Credal C4.5 with Refinement of Parameters

TitleCredal C4.5 with Refinement of Parameters
Publication TypeConference Paper
Year of Publication2018
AuthorsMantas, Carlos J., Abellán Joaquín, Castellano Javier G., Cano José R., and Moral Serafín
EditorMedina, Jesús, Ojeda-Aciego Manuel, Verdegay José Luis, Perfilieva Irina, Bouchon-Meunier Bernadette, and Yager Ronald R.
Conference NameInformation Processing and Management of Uncertainty in Knowledge-Based Systems. Applications
Pagination739–747
PublisherSpringer International Publishing
Conference LocationCham
ISBN Number978-3-319-91479-4
Abstract

Recently, a classification method called Credal C4.5 (CC4.5) has been presented which combines imprecise probabilities and the C4.5 algorithm. The action of the CC4.5 algorithm depends on a parameter s. In previous works, it has been shown that this parameter has relation with the degree of overfitting of the model. The noise level of a data set can influence on the choice of a good value for s. In this paper, it is presented a new method based on the CC4.5 method with a refining of its parameter in the time of training. The new method has an equivalent performance than CC4.5 with the best value of s for each level noise.