Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Eduard Manziuk has a total of 70 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    МЕТОД АНАЛІЗУ ТОНАЛЬНОСТІ ВІДГУКІВ КЛІЄНТІВ НА ОСНОВІ ГІБРИДНИХ МОДЕЛЕЙ КЛАСИФІКАЦІЇ ДЛЯ АВТОМАТИЗАЦІЇ ЗВОРОТНОГО ЗВ’ЯЗКУ В УПРАВЛІННІ ІТ-ПРОЄКТАМИ

    by Богдан Загурський, Едуард Манзюк, Тетяна Скрипник, Павло Радюк

    У статті представлено підхід до автоматизації процесів обробки зворотного зв’язку в системах управління ІТ-проєктами. Ефективний моніторинг задоволеності користувачів є критичним для забезпечення якості програмних продуктів, проте ручний аналіз великих масивів даних є трудомістким і неточним. У роботі запропоновано метод аналізу тональності відгуків клієнтів на основі гібридних моделей класифікації, що поєднує переваги класичних алгоритмів та глибокого навчання. Основу методу становить гібридна архітектура, що інтегрує SVM-класифікатор з RBF-ядром та глибоку нейронну мережу типу Dense Neural Network. Використання ймовірнісних оцінок SVM як додаткових вхідних ознак для нейронної мережі забезпечує ефективну корекцію помилок класифікації ("stacking"), а включення біграм дозволяє враховувати локальний контекст тексту. Система реалізована у вигляді модульної архітектури, що включає блоки векторизації та класифікації, забезпечуючи гнучкість інтеграції в існуючі інструменти управління проєктами та можливість масштабування. Експериментальні дослідження на збалансованому датасеті засвідчили високу ефективність запропонованого гібридного підходу: SVM досягнув точності 86.2%, нейронна мережа – 88.9%, а гібридна модель – 89.0% із середнім F1-Score 0.890. Комплементарність компонентів забезпечила стабільну перевагу гібридної моделі над індивідуальними алгоритмами. Отримані результати підтверджують доцільність використання розробленого методу для автоматизації зворотного зв’язку, що дозволяє менеджерам ІТ-проєктів оперативно отримувати об’єктивні метрики якості продукту та приймати обґрунтовані управлінські рішення.

  • Open Access

    Review of: "Certificate-Gated Clearance in Change-Impact Analysis: A Fail-Closed Architecture"

    by Eduard Manziuk

    The article addresses the critical regulatory compliance problem of risk asymmetry when assessing the impact of changes: a false escalation only costs analyst time, while a false clearance imperceptibly removes an obligation from the review queue. T...

  • Open Access

    МЕТОД ВИЯВЛЕННЯ ШАХРАЙСЬКИХ БАНКІВСЬКИХ ОПЕРАЦІЙ НА ОСНОВІ ТЕХНОЛОГІЙ ЗАСТОСУВАННЯ ЗАСОБІВ МАШИННОГО НАВЧАННЯ ТА ЗАХИСТУ ІНФОРМАЦІЇ В РОЗПОДІЛЕНО-ПАРАЛЕЛЬНОМУ ВИКОРИСТАННІ У ПРОЦЕСІ УПРАВЛІННЯ ІТ-ПРОЄКТАМИ

    by Владислав Ільчишин, Едуард Манзюк, Тетяна Скрипник, Руслан Багрій

    The rapid growth of digital financial transactions has created a complex environment for ensuring banking security. Detecting fraudulent transactions is increasingly challenging due to the diversity and evolving nature of fraud schemes. Traditional monitoring approaches based on manual analysis and simple rule-based systems are inefficient because of high labor costs, numerous false positives, and limited adaptability to emerging types of fraud. Advances in machine learning enable the development of adaptive fraud detection systems capable of automatically identifying complex patterns of suspicious behavior. Ensemble learning methods and techniques for handling class imbalance significantly improve transaction classification accuracy while reducing the workload on security services. The scarcity of labeled fraudulent examples limits conventional supervised models, and synthetic data generation methods such as SMOTE can create additional training samples, enhancing model performance. This study presents a method for detecting and classifying fraudulent banking transactions using a Random Forest classifier enhanced with SMOTE and integrated into a distributed-parallel processing framework suitable for IT project management. The approach includes modular data processing, model training, validation, and testing, providing scalability and adaptability. Experimental evaluation on a real-world bank transaction dataset shows high effectiveness: F1-score 0.83, precision 0.86, recall 0.79, and area under the ROC curve 0.93, outperforming logistic regression, decision trees, and standard Random Forest. The method detects approximately 79% of fraudulent transactions while maintaining a low false positive rate of 4.5%, demonstrating practical applicability in operational banking environments. The research contributes scientifically by advancing machine learning algorithms for anomaly detection in highly imbalanced financial data and developing adaptive detection systems. Practically, it reduces financial losses, improves transaction security, and increases the efficiency of bank security operations, enhancing trust in digital financial services. The proposed method provides a robust foundation for the development of intelligent and adaptive fraud detection systems in the financial sector.

  • Open Access

    Topography-Aware Deep Reinforcement Learning with Contextual Reward Engineering for Sustainable and Efficient Urban Traffic Control

    by Oleksander Ryzhanskyi, Oleksander Barmak, Eduard Manziuk, Pavlo Radiuk, Iurii Krak

    Urban traffic signal control heavily impacts vehicle emissions, yet most reinforcement learning models falsely assume flat terrain, ignoring the energy penalties of uphill stop-and-go driving. This omission creates a structural misalignment between generic, delay-focused rewards and the energetic realities of hilly corridors. In this work, we propose a topography-aware deep reinforcement learning framework that mitigates this hidden ecological cost. Our Context-Specific Reward Design procedure selects, normalizes, and calibrates reward terms based on physical conditions and traffic composition. The controller was trained using a microscopic simulation calibrated from video-derived traffic data, featuring a 3.8-degree uphill approach, 14,800 vehicles over 9 h, and a 20% heavy-vehicle fleet. In the uphill setting, the specialized controller reduced total CO2 emissions to 256.97 million milligrams, corresponding to 8.6% and 4.7% reductions relative to a pressure-based and a standard deep Q-learning controller, respectively. The proposed method also achieved the lowest mean trip duration of 72.09 s and a queue length of 1.31 vehicles. Welch’s t-tests confirmed that these CO2, duration, and queue improvements were significant. Overall, treating topography as a foundational design variable is crucial for sustainable urban mobility.

  • Open Access

    МЕТОД КЛАСИФІКАЦІЇ ПАТОЛОГІЙ ЛИСТЯ РОСЛИН НА ОСНОВІ ЗГОРТКОВИХ НЕЙРОННИХ МЕРЕЖ ІЗ ЗАСТОСУВАННЯМ ТЕХНОЛОГІЙ РОЗПОДІЛЕНОГО ПАРАЛЕЛЬНОГО НАВЧАННЯ

    by Валентин Соколовський, Едуард Манзюк, Руслан Багрій, Тетяна Скрипник

    The effective operation of automated plant disease diagnosis systems is a critical factor in modern precision agriculture, requiring robust solutions for early detection of pathologies. Traditional diagnostic methods based on visual inspection are labor-intensive, subjective, and difficult to scale. This paper presents an improved method for classifying pathologies of agricultural plant leaves based on deep learning technologies, specifically aimed at increasing diagnostic accuracy and optimizing computational performance in high-load environments. A modified five-block Convolutional Neural Network (CNN) architecture, derived from the VGG16 baseline, is proposed. The key architectural innovation involves the deep integration of batch normalization mechanisms after convolutional layers and dropout regularization in the fully connected layers. These modifications successfully addressed the issue of overfitting on limited datasets, ensuring the model's robustness to variations in input data and improving feature extraction capabilities for complex disease patterns. To ensure the efficiency of experimental studies involving large-scale image datasets, a distributed parallel training technology was implemented. This approach relies on the principle of data parallelism with synchronous gradient updates across multiple computing nodes. The implementation allowed for a significant reduction in model training time and provided horizontal scalability of the system, making it suitable for processing big data. Furthermore, the paper describes the technological aspects of software creation, emphasizing the use of declarative configuration and a comprehensive versioning system (following MLOps principles). This approach guarantees the full reproducibility of experiments, systematic documentation of the development process, and reliability of the obtained results. Experimental studies were conducted using a representative dataset of bean leaf images classified into four distinct categories. The results established that the proposed method achieves a classification accuracy of 91.2%, outperforming the baseline model by 4%. The study also proved the critical impact of data augmentation techniques on the model's generalization ability, particularly under conditions of variable lighting and diverse shooting angles. The obtained results confirm the practical value of the method for developing scalable automated diagnostic systems.

  • Open Access

    АВТОМАТИЗОВАНЕ ВІДНОВЛЕННЯ ТРАСУВАЛЬНИХ ЗВ’ЯЗКІВ МІЖ ВИМОГАМИ ТА ПРОГРАМНИМ КОДОМ З ВИКОРИСТАННЯМ ВЕЛИКИХ МОВНИХ МОДЕЛЕЙ

    by Олександр Скрипнюк, Руслан Багрій, Едуард Манзюк, Тетяна Скрипник

    The problem of consistency between software requirements and their implementation in source code becomes critically important as the scale and complexity of modern software systems increase, since the absence of reliable traceability links often leads to incomplete requirement implementation, complicates code maintenance, and hinders the verification of system correctness. Manual generation of traceability matrices is a labour-intensive and error-prone process, especially in large projects. The use of large language models opens up new opportunities for automating this process, as such models are capable of reflecting deep semantic relationships between text requirements and software code fragments. The article proposes a method for identifying traceability links between requirements and software code using transformer models of large language systems. The proposed approach is based on converting text artefacts into vector representations using CodeBERT, SBERT, and TF-IDF models, followed by calculating semantic similarity to automatically identify potential connections. The method covers such stages as data preparation, embedding generation, search for relevant fragments, and evaluation of the results obtained. The experiments were conducted on the MSR-2021 dataset, which contains real traceability links for several projects. The obtained results demonstrated the advantage of CodeBERT over traditional approaches (TF-IDF, SBERT): the method achieves accuracy of up to 0.85 and an F1-score of up to 0.50, depending on the search depth, which represents strong performance for automated information retrieval and ranking tasks. The study additionally confirmed the importance of considering the structural context of the code and showed the influence of the Top-K parameter on the balance between recall and precision. The results indicate that integrating LLM-based models significantly improves the level of automation, consistency, and quality of requirements traceability in modern software development environments.

  • Open Access

    Method of adaptive knowledge distillation from multi-teacher to student deep learning models

    by Oleksandr Chaban, Eduard Manziuk, Pavlo Radiuk

    Transferring knowledge from multiple teacher models to a compact student model is often hindered by domain shifts between datasets and a scarcity of labeled target data, degrading performance. While existing methods address parts of this problem, a unified framework is lacking. In this work, we improve multi-teacher knowledge distillation by developing a holistic framework, enhanced multi-teacher knowledge distillation (EMTKD), that synergistically integrates three components: domain adaptation within teacher training, an instance-specific adaptive weighting mechanism for knowledge fusion, and semi-supervised learning to leverage unlabeled data. On a challenging cross-domain cardiac MRI benchmark, EMTKD achieves a target domain accuracy of 88.5% and an area under the curve of 92.5%, outperforming state-of-the-art techniques by up to 5.0%. Our results demonstrate that this integrated, adaptive approach yields significantly more robust and accurate student models, enabling effective deep learning deployment in data-scarce environments.


  • Open Access

    An Adaptive Machine Learning Approach to Sustainable Traffic Planning: High-Fidelity Pattern Recognition in Smart Transportation Systems

    by Vitaliy Pavlyshyn, Eduard Manziuk, Oleksander Barmak, Pavlo Radiuk, Iurii Krak

    Effective and sustainable planning for future smart transportation systems is hindered by outdated traffic management models that fail to capture real-world dynamics, leading to congestion and significant environmental impact. To address this, advanced machine learning models are required to provide high-fidelity insights into urban mobility. In this work, we propose an adaptive machine learning approach to traffic pattern recognition that synergizes the HDBSCAN and k-means clustering algorithms. By employing a data-driven weighted voting mechanism, our solution provides a robust analytical foundation for sustainable planning, integrating structural analysis with precise cluster refinement. The crafted model was validated using a high-fidelity simulation of the Khmelnytskyi, Ukraine, transport network, where it demonstrated a superior ability to identify distinct traffic modes, achieving a V-measure of 0.79–0.82 and improving cluster compactness by 10–14% over standalone algorithms. It also attained a scenario identification accuracy of 92.8–95.0% with a temporal coherence of 0.94. These findings confirm that our adaptive approach is a foundational technology for intelligent transport systems, enabling the planning and deployment of more responsive, efficient, and sustainable urban mobility solutions.


  • Open Access

    МЕТОД ГЕНЕРАЦІЇ БАГАТОТРЕКОВИХ СИМВОЛІЧНИХ КОМПОЗИЦІЙ ЗА ДОПОМОГОЮ ГЕНЕРАТИВНОГО ШТУЧНОГО ІНТЕЛЕКТУ

    by Андрій Прийма, Едуард Манзюк, Олександр Пасічник, Тетяна Скрипник

    This study explores a method for symbolic music generation using artificial intelligence techniques. Unlike traditional rule-based or statistical approaches, the proposed method employs deep generative models to automatically create multi-track, polyphonic musical compositions. As part of the research, a comparative experiment was conducted to assess the quality of music generation by different architectural variants of the models. The main task of this stage was to determine the ability of the models to reproduce key characteristics of musical logic. The focus is placed on capturing the temporal structure of music and ensuring harmonic and rhythmic coherence across multiple instrument tracks. A model architecture is introduced that generates music in the form of piano-roll representations, treating musical bars as fundamental units. The system supports both autonomous generation and human-AI collaboration, where the model can accompany a user-provided melody with additional tracks. A specially prepared dataset of symbolic music was used for training, and a series of objective metrics were developed to evaluate the quality of the generated output. The proposed architecture ensures the formation of coherent musical phrases while preserving rhythm, harmony, and inter-track interaction. Analysis of intermediate results showed a gradual increase in the quality of generation: from chaotic sounds to ordered structures, which indicates the effectiveness of training. Results from both quantitative analysis and a subjective user study demonstrate that the method is capable of producing musically coherent and structurally consistent compositions. This research contributes to the development of AI-assisted tools for creative music production and expands the possibilities of automated composition in the symbolic domain.

  • Open Access

    МЕТОД КЛАСИФІКАЦІЇ КОНФІДЕНЦІЙНОЇ ІНФОРМАЦІЇ ІЗ ЗАСТОСУВАННЯМ МАШИННОГО НАВЧАННЯ

    by Богдан Палійчук, Едуард Манзюк, Тетяна Скрипник, Олександ Пасічник

    This article focuses on the development and improvement of a confidential information classification method based on text data analysis using machine learning. The primary goal of the research was to create an effective tool for automatic detection of sensitive information in structured and unstructured data to enhance their protection. The method employs machine learning algorithms, specifically the Naive Bayes classifier with Laplace smoothing, and Support Vector Machine (SVM) for comparative evaluation. The dataset used for training and testing included real corporate data, public datasets, and synthetic examples, ensuring high diversity and representativeness. The developed method demonstrates high accuracy (92%), recall (90%), and F1-score (91%), outperforming traditional approaches such as Naive Bayes (84% accuracy) and SVM (88% accuracy). Special attention was given to handling rare or unique data classes by applying Laplace smoothing, which significantly improved the model’s robustness and stability. This approach enables more reliable identification of confidential data, which is critical for ensuring information security in organizations. The article also details the data preparation process, including cleaning, tokenization, normalization, and splitting into training, validation, and test sets. A comparative analysis of different algorithms’ effectiveness was conducted, with results presented using accuracy, recall, and F1-score metrics. Recommendations for further improvements include expanding functionality, enhancing scalability, and adapting to specific industry requirements. The results confirm the promise of machine learning for automated protection of confidential information and can be useful for security system developers, information security researchers, and practitioners involved in data protection. The proposed method has potential for integration into corporate information management systems and contributes to improving the overall level of cybersecurity.