Abstract
This paper presents a new approach in machine learning, especially, in supervised classification and reasoning under uncertainty. For many classification problems, uncertainty is often inherent in modeling applications and should be treated carefully and not rejected in order to make better decisions. Artificial Immune Recognition System (AIRS) is a well known classifier that has provided good results in the certain context. However, this method is not able to cope with uncertainty. In order to overcome this limitation, we propose a new classification approach combining the AIRS and possibility theory. The new approach is allowing to deal with uncertain attribute values of training instances. The uncertainty is expressed via possibility distributions. Experimentations on real data sets from the U.C.I machine learning repository show good performances of the proposed approach.