МЕТОД ВИЗНАЧЕННЯ ВИКОНАННЯ УМОВ ФУНКЦІОНАЛЬНОЇ СТІЙКОСТІ РОЗПОДІЛЕНИХ ІНФОРМАЦІЙНИХ СИСТЕМ ЗА ТОПОЛОГІЧНОЮ СТРУКТУРОЮ ІЗ ЗАСТОСУВАННЯМ МАШИННОГО НАВЧАННЯ

<jats:p>The subject of this research is the application of machine learning to detect the functional stability of distributed information systems based on the numerical characteristics of their topology. This study aims to improve the accuracy of verifying the condition whose fulfillment ensures the functional stability of considered distributed information systems. The task of this research is to improve the accuracy of machine learning-based verification of the condition whose fulfillment ensures the functional stability of a considered distributed information system. The main result of the study is the development of an ensemble classifier based on multiple heterogeneous machine learning models, enabling superior accuracy compared to that reported in related studies on the same or similar tasks. Unlike other studies on applying machine learning to assess information system functional stability, the proposed classifier architecture allows sequential use of sets of machine learning models. In other words, the first set of models receives a group of input parameters and estimates certain additional parameters. The obtained estimates are then passed to another set of models, which produces the final result. This approach enables the discovery of deeper dependencies between input and output parameters, which in turn yields higher accuracy. Such an approach differs significantly from the one commonly used in practice, where all models receive the same input parameters, produce the same type of output, and aggregate their outputs depending on the specific task. Conclusions. The study's outcome is a trained and tested ensemble machine learning model that, based on several numerical characteristics, detects whether the condition is met for the considered distributed information system to be regarded as functionally stable. The ensemble machine learning model obtained in this work can be applied at the design stage of information systems. When used as demonstrated here, the machine learning models trained in the study can be integrated into software for the design and simulation of information systems. The method for combining these machine learning models is of particular interest. As described in the paper, the approach to organizing models within the ensemble opens new possibilities for building highly accurate classifiers. The principal scientific novelty of the obtained results lies in a new approach to constructing ensemble machine learning models. Unlike classical ensembles, such as random forests, the ensemble described in this study employs a set of machine learning models to compute estimates of certain parameters, which are then used as inputs to another set of models. This particular approach to organizing the ensemble most likely improved the quality of solving the task, as demonstrated within the study.                                                              </jats:p>