A possibilistic classification approach to handle continuous data

Naive Possibilistic Network Classifiers (NPNC) have been recently used to accomplish the classification task in presence of uncertainty. Because they are mainly based on possibility theory, they run into problems when they are faced with imperfection where the possibility theory is the most convenient tool to represent it. In this paper we investigate to develop a new classification approach for perfect/imperfect (imprecise) continuous attribute values under the possibilistic framework based mainly on Possibilistic Networks. To build the naive possibilistic network classifier, we develop a procedure able to deal with perfect or imperfect dataset attributes which is used to classify new instances that may be characterized by imperfect attributes. We have tested our approach on several different datasets. The results show that this approach is efficient in the imperfect case.