Handling Uncertain Attribute Values In Decision Tree Classifier Using The Belief Function Theory

Asma Trabelsi (1) , Zied Elouedi (2) , Eric Lefevre (3)

1 Expleo Group
2 LARODEC - Laboratoire de Recherche Opérationnelle de Décision et de Contrôle de Processus
3 LGI2A - Laboratoire de Génie Informatique et d'Automatique de l'Artois

Decision trees are regarded as convenient machine learning techniques for solving complex classification problems. However, the major shortcoming of the standard decision tree algorithms is their unability to deal with uncertain environment. In view of this, belief decision trees have been introduced to cope with the case of uncertainty present in class' value and represented within the belief function framework. Since in various real data applications, uncertainty may also appear in attribute values, we propose to develop in this paper another version of decision trees in a belief function context to handle the case of uncertainty present only in attribute values for both construction and classification phases.

Domaines: Intelligence artificielle [cs.AI]

Asma Trabelsi, Zied Elouedi, Eric Lefevre. Handling Uncertain Attribute Values In Decision Tree Classifier Using The Belief Function Theory. International Conference on Artificial Intelligence: Methodology, Systems, and Applications, AIMSA'2016, Sep 2016, Varna, Bulgaria. pp.26-35, ⟨10.1007/978-3-319-44748-3_3⟩. ⟨hal-03649449⟩