M. Ramoni, A. Riva, M. Stefanelli, and V. Patel
Bayesian Belief Networks (BBNs) provide suitable formalism for many medical applications. Unfortunately, learning a BBN from a database is a very complex task, since the specification of a BBN requires a large amount of information that is not always available in real-world databases. In this paper, we introduce a new class of BBNS, called Ignorant Belief Networks, able to reason with incomplete information. We will show how this new formalism can be used to forecast blood glucose concentration in insulin-dependent diabetic patients using underspecified probabilistic models directly derived from a real-world database.