In the context of developing modern intel-ligent information systems, one of the key tasks is to build models that can effectively work with incomplete, fuzzy or uncertain data. Predictive modelling often faces the problem of the lack of complete information about objects or processes, which complicates the establishment of reliable analytical conclusions. In such cases, traditional statistical methods demonstrate limited flexibility, while probabilistic approaches, particularly Bayesian networks, allow taking into account uncertainty and partial information.
A special place among Bayesian modelling tools is occupied by nodes of the Noisy-MAX type, which allow compactly displaying the dependencies between a set of causes and a single effect in multi-valued discrete systems. The use of such nodes allows significantly reducing the computational complexity of models and ensuring resistance to data gaps. This makes this approach especially relevant for the tasks of medical diagnostics, forecasting technical failures, and assessing risks in financial and environmental systems.
The paper considers the sequence of processing incomplete data in the predictive modelling using Noisy-MAX nodes, which increases the accuracy and reliability of forecasts under conditions of input information uncer-tainty. To achieve this goal, the basic principles of con-structing such nodes, probability updating algorithms, and practical aspects of their application in forecasting prob-lems are considered.