Possibilistic induction in decision-tree learning

We propose a generalization of Ockham's razor, a widely applied principle of inductive inference. This generalization intends to capture the aspect of uncertainty involved in inductive reasoning. To this end, Ockham's razor is formalized within the framework of possibility theory: It is not simply used for identifying a single, apparently optimal model, but rather for concluding on the possibility of various candidate models. The possibilistic version of Ockham's razor is applied to (lazy) decision tree learning.