Laboratoire de Génie Informatique et d’Automatique de l’Artois

In the complex pattern classification problem, how to well deal with the uncertainty remains an interesting and important research topic. Belief function theory provides an efficient tool to represent and combine the uncertain information, and it is employed here for pattern classification. In this talk, we will present three parts of our recent work on the belief-based pattern classification :

1. credal classification of incomplete pattern with missing attributes
2. credal c-means clustering method for the partly overlapping data and
3. classifier fusion with refined reliability evaluation under the power-set.

We will also discuss some potential research work with belief functions in the final.