У дослідженні був запропонований новий метод побудови набору важливих ознак для вирішення задач класифікації. Цей метод заснований на ідеє використання ансамбля оцінювачів важливості ознак з підведенням підсумків і кінцевого результату ансамбля за допомо-гою алгоритмів нечіткої логіки. В якості оцінювачів важливості ознак було використано статистичні критерії (chi2, f_classif, коефіцієнт кореляції), критерій середнього зменшення помилок класифікації (mean decrease in impurity - MDI), критерій взаємної інформації (mutual_info_classif). Зменшення кількості ознак на усіх наборах даних впливає на точність оцінювання відповідно до критерію середнього зменшення помилок класифікації. Поки група ознак в на-борі даних для навчання містить перши за списком ознаки з найбільшим впливом, точність моделі знаходиться на початковому рівні, але при виключенні з моделі хоча б однієї з ознак з великим впливом, точність моделі помітно знижується. Найкращі результати класифікації для усіх досліджених наборів даних забезпечили класифікатори на основі дерев або найближчих сусідів: DecignTreeClassifier, ExtraTreeClassifier, KNeighborsClassifier. За рахунок виключення із моделі несуттєвих ознак досягається помітне збільшення швидкості навчання (до 60-70%). Для підвищення точності оцінювання було використано ансамблеве навчання. Найкращі показники за швидкістю навчання забезпечив класифікатор VotingClassifier, побудований на базі алгоритмів з максимальною швидкістю навчання. Для майбутньої роботи метою є подальше вдосконалення запропонованої моделі IDS в напрямках вдосконалення вибору класифікаторів для отримання оптимальних результатів, та налаштування параметрів вибраних класифікаторів, удосконалення стратегії узагальнення результатів окремих класифікаторів. Для запропонованої моделі істотний інтерес представляє можливість виявлення окремих типів атак з урахуванням багатокласового прогнозування.
1 January 2019 in Cybersecurity: Education, Science, Technique
Sensor networks are one of the most relevant and promising technologies for wide application in various spheres of human life. Relatively inexpensive components, namely, sensor nodes are combined into one network. Due to the use of a significant number of nodes, in addition to the overall impact on functionality, this also leads to a decrease in the reliability of the network. Wireless connectivity has a number of limitations when it comes to the direct connection to the public communications network or the next-generation networks with optical components. For example, the relatively small distance to send information between constituent components. It follows next idea -sooner or later there will be the denial of nodes, which will lead to the isolation of other sensors. To avoid this, or at least to ensure the network connectivity during a failover, more nodes on a separate site need to be used. Also, due to the presence of connectivity between touch nodes, there are a number of general features that should be taken into account when designing such a network and deploying it in real-world conditions. One of these tasks is to ensure the accuracy of packets of information’s transfer between the sensor nodes, as violations of its integrity, quality, reliability and safety can lead to serious consequences. Due to insignificant volumes of memory, sensor nodes are not able to capture data about all nodes, their physical addresses, metrics for the fastest transmission of information packets. In view of this, the number of requests for the same node may exceed the critical rate at a certain time. However, there may also be a situation where a node that has not been earlier tested can be connected to the network. In this case, the question of the authenticity of the data that will circulate in the network from a similar, hidden node to others is sharply raised. Therefore, it is necessary to study the existing and the introduction of new algorithms for the transmission of information in the presence of hidden nodes in wireless sensory networks at the present stage of development.
Artificial intelligence (AI)-related harms are increasingly attributed to governance failures rather than to isolated technical malfunctions. This article reframes AI governance as a core managerial competence grounded in leadership authority, accountability design, and organizational communication. The study addresses a persistent gap in higher education and managerial training, namely the insufficient preparation of future leaders to govern AI-mediated decision systems responsibly. Using a structured conceptual synthesis grounded in socio-technical systems theory and the organizational governance literature, the paper identifies recurring governance failure modes, including authority drift from human decision-makers to automated systems, diffusion of accountability, governance debt accumulation, and reliance on average-case performance metrics that obscure worst-case risks. To illustrate early governance readiness, an exploratory survey of senior university students—representing early-stage managerial cohorts—was conducted, resulting in the AI Governance Readiness Composite Score (AGRCS). The findings illustrate preliminary patterns in self-assessed governance readiness among early-stage managerial cohorts, without implying statistical generalization or population-level conclusions. The study does not seek statistical generalization but uses empirical signals to support conceptual arguments. The main contribution lies in positioning leadership authority, intervention capacity, and governance-related communication as central pillars of sustainable AI governance. The article translates these governance principles into an educational agenda, proposing sustainable pedagogy practices such as authority mapping, escalation rehearsals, worst-case simulations, and governance-focused learning environments. By framing AI governance as a leadership and communication challenge rather than a narrow technical problem, the study contributes to sustainable organizational development, responsible decision-making, and long-term societal trust aligned with the United Nations Sustainable Development Goals.
As a result of a combination of template comparison for estimating block similarity, edge selection algorithms and an improved block search method, two versions of a generalized method for finding motion vectors, called TFEMVFAS (Three by Frame Edge Map Motion Vector Field Adaptive Search), were proposed. Both modifications of the method were tested according to the comprehensive testing recommendations of the international commission MPEG for the evaluation of complex video tools, on the example of 13 video sequences, with different bitrates. The effect of increasing the size of the blocks on which the image is divided on the PSNR value and the maximum acceleration was also investigated.
27 June 2025 in Informatyka, Automatyka, Pomiary w Gospodarce i Ochronie Środowiska
The article is devoted to modern software systems (SS) and improving their quality using machine learning methods, including the Integrated Gradients (IG) method. Key problems and limitation of IG use in real operating conditions of the SS, such as complexity of systems, correlation of variables and computing efficiency are considered. Ways to improve IG, including adaptive integration, spatial smoothing and use of weight factors, are proposed. Experimental results are described that confirm the effectiveness of the proposed modifications to improve the quality of the SS. Adaptive integration has achieved the best results (MAE 0.11), adaptability and interpretation.
18 June 2026 in Information and Telecommunication Sciences
Background. The rapid growth of telemetry and event logs in corporate and public-sector information and communication systems complicates the formation of reliable incidents in SIEM platforms. Traditional correlation methods that rely on fixed rules, temporal windows, and heuristic similarity standards fall short in the face of evolving threats and fail to consider the risk context associated with assets. In addition, correlation is vulnerable to log manipulation (log injection, partial context modification, timestamp distortion), which can generate false incidents or conceal multi-stage attacks and overload SOC operations. Objective. This paper aims to develop a cryptographically resilient SIEM event-correlation method based on risk-oriented assessment and context-hashed similarity, improving the trustworthiness of incident formation and ensuring robustness against injection, substitution, and partial compromise of event logs. Methods. We formalise a security event as a structured object comprising context, temporal parameters, and risk, and propose normalisation and canonicalisation procedures for context features with a fixed ordering. To protect context, we employ a context-hashed representation with multi-level fingerprints . Event similarity is defined via a cryptographically resilient fingerprint-matching metric, while correlation is computed using an integral indicator Corr that combines contextual similarity with the risk significance of events. A time-window correlation algorithm with a threshold-based decision rule is constructed. Results. The proposed method combines cryptographically protected context similarity with risk-oriented assessment, providing resilience to context and log manipulation. We develop a threat model for SIEM alerts as correlation objects and a mapping table “attack class → impact → mitigation mechanism.” Simulation experiments in a corporate environment confirm a reduced false-incident rate and improved correlation precision while maintaining comparable incident-detection recall; robustness is demonstrated under log injection, partial tampering, and timestamp shift scenarios. In practice, the method can be integrated as a correlation module into existing SIEM platforms without substantial architectural redesign. Conclusions. Integrating context-hashed representations with risk-oriented correlation yields a cryptographically resilient mechanism for automated incident formation in SIEM, mitigates alert flooding, and increases SOC trust in correlation outcomes under dynamic threats. Future work should focus on adaptive tuning of α and θ using SOC feedback and ML-based models, as well as expanding experimental validation on real industrial data.