Development of a method to improve the reliability of assessing the condition of the monitoring object in special-purpose information systems

• Published in: Eastern-European journal of enterprise technologies Vol. 2; no. 3 (116); pp. 6 - 14
• Main Authors: Sova, Oleg, Radzivilov, Hryhorii, Shyshatskyi, Andrii, Shvets, Pavel, Tkachenko, Valentyna, Nevhad, Serhii, Zhuk, Oleksandr, Kravchenko, Serhii, Molodetskyi, Bohdan, Miahkykh, Hennadii
• Format: Journal Article
• Language: English
• Published: 28.04.2022
• ISSN: 1729-3774; 1729-4061
• DOI: 10.15587/1729-4061.2022.254122

The peculiarities of modern military conflicts significantly increase the requirements for the efficiency of object state assessment. Therefore, it is necessary to develop algorithms (methods and techniques) that can assess the state of the monitoring object from different sources of intelligence for a limited time and with a high degree of reliability. Accurate and objective object analysis requires multi-parameter estimation with significant computational costs. That is why the following tasks were solved in the study: the formalization of the assessment of monitoring objects was carried out, a method of increasing the efficiency of assessing the condition of monitoring objects was developed and an efficiency assessment was carried out. The essence of the proposed method is the hierarchical hybridization of binary classifiers and their subsequent training. The method has the following sequence of actions: determining the degree of uncertainty, constructing a classifier tree, determining belonging to a particular class, determining object parameters, pre-processing data about the object of analysis and hierarchical traversal of the tree. The novelty of the method lies in taking into account the type of uncertainty and noise of the data and taking into account the available computing resources of the object state analysis system. The novelty of the method also lies in the use of combined training procedures (lazy training and training procedure for evolving neural networks) and selective use of system resources by connecting only the necessary types of detectors. The method allows you to build a top-level classifier using various low-level schemes for combining them and aggregating compositions. The method increases the efficiency of data processing by 12–20 % using additional advanced procedures