Vitko O. Multilevel probabilistic networks for modelling com-plex information systems with uncertainty

This work is devoted to development of Bayesian probabilistic metanetworks and their learning methods for analysis and modelling of complex information systems with uncertainty. The new architecture of multilevel probabilistic metanetwork is developed. Based on the architecture three models are developed: C-metanetwork for management of conditional dependencies, R-metanetwork for management of feature selection and combined RC-metanetwork, which differ by the way of context influence. The rules for probabilistic inference in Bayesian metanetwork are modified. The learning methods are extended for multilevel models. The use of probabilistic metanetworks for several actual tasks of modelling of information systems is illustrated.