STSvitlana Terenchuk

Svitlana Terenchuk

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Svitlana Terenchuk has a total of 42 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Reducing the complexity of gradient boosting models for assessment of the technical condition of real estate objects

    by Oleksii Yehorov, Anatolii Pashko, Bohdan Yeremenko, Svitlana Terenchuk

    This study is aimed at developing a tool that provides systematic support for expert decisions in assessing of the technical condition of real estate objects in conditions of combat operations and post-war reconstruction. The research focuses on improving the interpretability of machine learning models used in the decision support systems. It approach introduces synthetic features representing distances to cluster centroids obtained through unsupervised learning, allowing latent structural patterns in the data to be captured. It enrich the feature space and simplify the decision boundaries learned by the gradient boosted decision trees model. Experimental results obtained on the test dataset demonstrate that the proposed approach allows to preserve predictive while reducing model complexity. In particular, the number of estimators required by the gradient boosted decision trees model was reduced drastically while maintaining the same classification accuracy. Experiments analyzing the influence of the number of clusters on model complexity confirm the effectiveness of the proposed feature augmentation strategy. The scientific novelty lies in the development of a feature space expansion method based on clustering, which combines unsupervised learning and gradient boosting to reduce the number of weak models required to achieve a given prediction accuracy. Unlike traditional approaches, which focus on maximizing accuracy, this method aims to achieve a balance between model complexity and interpretability, which is critically important for expert-driven decision support systems. The practical significance of the work lies in improving model interpretability by reducing the number of weak learners required for accurate predictions.

  • Open Access

    Development of an approach to managing transport logistics at a construction site

    by Pavlo Pasieka, Oleksii Panko, Nataliia Poltorachenko, Svitlana Terenchuk, Bohdan Yeremenko

    This study investigates the process that forms a vehicle route in a complex logistics environment. The task addressed relates to the lack of a full-fledged logic in autonomous decision-making regarding operational routing in specialized logistics systems. An analysis of modern transport logistics management systems revealed their limitations in terms of the efficiency and explainability of the logic of autonomous decision-making regarding route replanning. An approach to route formation has been devised that combines the modified A* algorithm, multi-agent reinforcement learning, and spatio-temporal graph convolutional networks (STGCNs). Distinctive features of this approach are the combination of spatial-temporal forecasting, multi-agent decision-making, and modular system architecture, which provides adaptability, scalability, resistance to communication degradation, as well as explainability of transport routing decisions in closed reconfigurable networks. Experimental comparison of the devised approach and analogs based on A* and STGCN, carried out on a simulation model of the logistics environment at a construction site in residential area, showed its statistical superiority in terms of the route travel time criterion. With a threshold of this criterion of 300 s, the proposed approach achieved this indicator in 90% of cases, and the approaches based on STGCN and A* – in 79% and 57% of cases. In this case, the value of the 95th percentile was 310 s, 360 s, and 410 s for the proposed approach, STGCN, and A*. These results are explained by the combination of network state prediction and adaptive route replanning. The findings could be used in the design of intelligent transport logistics management systems at other facilities with dynamic transport infrastructure


  • Open Access

    Model of route corridor formation based on predictive assessments by marl agents

    by Pavlo Pasieka, Svitlana Terenchuk

    The object of the study is the process of forming route decisions for vehicles in the dynamic environment of a construction site. The subject of the study is a model for coordinating partial assessments of MARL agents for the formation of a route corridor at the macrolevel of the spatial representation of the logistics environment. The problem being solved is the adaptive narrowing of the route search space under conditions of variable topology of passages, mutual influence of vehicles, incompleteness of observations, and the need for explainability of the final route decision. The purpose of the study is to develop a model for forming a vehicle route corridor on a construction site, which is based on expert assessments of MARL agents regarding the predicted state of the logistics environment, operational priorities of the construction process, availability of sections, and the risk of transport conflicts. The work uses methods of graph modeling, hierarchical hexagonal representation of the environment, multi-agent reinforcement learning, centralized training with decentralized execution (CTDE), and multi-criteria evaluation. A model is proposed in which functional MARL agents analyze the macrolayer of the hexagonal geospatial model of the movement space from different expert positions and form a preference map for building a route corridor. In contrast to approaches that use machine learning for the direct selection of a complete route, the proposed model forms a spatial constraint for subsequent classical planning, passing a limited subgraph of promising sections to the classical planner rather than the entire graph. The scientific novelty lies in the further development of models of intelligent formation of logistics routes through the introduction of expert interaction of MARL agents at the stage of building a route corridor. The practical significance of the results lies in the possibility of integrating the model into an event-driven information system for managing the transport logistics of a construction site to increase the adaptability, resilience, and explainability of routing.

  • Open Access

    Improvements to the circuit breaker prediction model for increasing the fault tolerance of microservices

    by Anatolii Pashko, Ihor Paprotskyi, Andrii Glybovets, Svitlana Terenchuk

    In high-load distributed systems, it is crucial to pay strict attention to the reliability of communication between architectural components, making fault tolerance a fundamental property of microservice architecture. While various fault tolerance patterns exist to improve microservice resilience, and their implementations are widely adopted, none are universal for the range of possible failures in service communication. Therefore, there is a continuous need to search for more effective approaches to achieving fault tolerance in distributed software systems. This research is building upon our earlier study, which introduced the Circuit Breaker Prediction Model. The main principle of the Prediction model is to calculate the rating of communication between two services, defined as a convex combination of various performance metrics, and the rating value is then used to determine whether the interaction with the target service can be safely re-established after the period of blocking outgoing requests. Compared to standard implementations of this method, it was specifically designed to reduce the latency involved in blocking request transmission between microservices, especially under uncertain load conditions. The purpose of this paper is to enhance this model and verify its effectiveness when combined with other resilience patterns, particularly Retry and Timeout. The study enhances the flexibility of the Circuit Breaker Prediction Model by describing an adaptable sliding window mechanism that dynamically adjusts the context size based on load conditions at runtime. Also, the algorithm of rating calculation is generalized to enable the usage of any substantial number of performance metrics. Finally, this research conducts simulated experiments with multiple service communication configurations and fault tolerance pattern combinations, using the standard Circuit Breaker implementation as a baseline for comparison. The findings show improved communication performance, mainly due to reduced request transmission delays under the same load conditions. The prospects of applying this model’s principles to other fault tolerance patterns have been discussed, with particular emphasis on the potential of using machine-assisted optimization techniques in dynamic decision making.

  • Open Access

    Concept of developing an integrated approach to the formation of logistic routes

    by Pavlo Pasieka, Svitlana Terenchuk

    The development of information systems for the formation of logistic routes used in the field of transport logistics is studied, and algorithmic approaches to solving the vehicle routing problem are analyzed. The main goal of this study is to define the concept of developing these systems toward improving their ability to function in dynamic and stochastic environments in real time. Classical algorithmic approaches to routing problems, such as the shortest path search, the traveling salesman problem, the transportation problem, and their modifications oriented towards real-time constraints, are considered. Intelligent and hybrid dynamic route planning systems using heuristic methods, machine learning, and reinforcement learning are investigated. It is shown that most existing routing systems use simplified or fragmentary environmental models, which limits their ability to adapt to significant changes in traffic conditions, infrastructure, or external factors. It is revealed that the key limitations of current approaches are the insufficient universality of environmental data representation, weak integration between analytical and intelligent components, and the lack of an expert level of decision coordination regarding the determination of the best route. The expediency of using graph models of the environment as a universal form of storage and processing of logistic data, suitable for both classical routing algorithms and artificial intelligence methods, is substantiated. A concept of an integrated approach to the formation of logistic routes is proposed, combining classical route search algorithms with intelligent environmental analysis methods and multi-agent systems. This concept creates the prerequisites for increasing the adaptability, resilience, and scalability of logistic route formation systems without abandoning proven algorithmic methods.

  • Open Access

    Improving of intelligent decision support systems for planning a balanced diet

    by Viktor Ladyzhets, Svitlana Terenchuk, Iryna Aznzurian, Antonina Makhynia

    The object of this study is the process of creating a personalized menu. The subject of the study is recommendation systems for generating breakfast, lunch, snack, and dinner menus. The task solved was the development of an effective system for supporting the decisions by a wide range of users in planning a balanced diet. To form a menu of dishes of different categories of meals in a hybrid system for planning a balanced human diet, it is proposed to use different recommendation systems based on different models of artificial intelligence. The choice of the singular matrix decomposition model, the gradient boosting model of decision trees, and the wide and deep learning models for recommendation systems for forming a menu of dishes has been substantiated by the results of analysis. Based on the results of the experiment with these artificial intelligence models, it was determined which of them are more effective in solving the problem of forming a menu of meals for different categories of meals. The effectiveness of all models was evaluated by such test indicators as Precision@K, mean absolute and root mean square error. The feasibility of choosing the singular matrix decomposition model for generating breakfast menus and the wide and deep learning models for generating snack, lunch, and dinner menus was evaluated by the Precision@K values. The singular matrix decomposition model, compared to the other models studied in this paper, showed the highest Precision@K for breakfast, namely 0.942. The wide and deep learning models demonstrated the highest Precision@K for lunch, snack, and dinner: 0.961, 0.977, and 0.951, respectively. In practice, the results could be used to develop highly efficient personalized meal planning services in mobile and online platforms

  • Open Access

    Support system real estate reconstruction process: processing and storage of data

    by Anatolii Buhrov, Bohdan Volokh, Ihor Bosenko, Svitlana Terenchuk, Анатолій Бугров, Богдан Волох, Ігор Босенко

    This article is a continuation of the work aimed at developing an information and communication system to support the process of restoring real estate objects that have been damaged and destroyed as a result of military operations conducted by the russian federation with the support of the republic of belarus on the territory of Ukraine since February 24, 2022. A holistic view of the highly loaded distributed architecture of a scalable and fault-tolerant system for supporting the process of restoring real estate objects, designed to solve many problems of the construction industry of Ukraine, has been formed. A universal model of the data processing and storage module is proposed. This module is designed for the rapid processing, analysis and storage of a large amount of information presented in various formats, taking into account the need to process fuzzy textual information from the media and eyewitnesses, and visual data obtained from geographic information systems, satellite surveillance systems and unmanned aerial vehicles. The article shows a diagram of the sequence of interaction between services and the importance of ontology in solving the problems of overload and avoiding conflicts in understanding data coming from different sources. Based on the analysis of a number of studies in the form of optimization ideas, recommendations for potential improvement of the system ontology formation are provided. At the same time, the possibility of using the module in the process of ontology formation and ways to improve ontology-managed systems are shown. The practical significance of the work is to ensure continuous monitoring and timely updating of data on real estate objects, which is important for maintaining the relevance of information when performing restoration work.


  • Open Access

    Designing the recommendation method for a meal planning information system

    by Viktor Ladyzhets, Svitlana Terenchuk

    The work is devoted to solving such a complex scientific and applied problem of dietetics as the formation of recipes for dishes of different food categories. The purpose of the article is to develop a method for recommending recipes based on machine learning, which allows taking into account the personal preferences and dietary restrictions of the user. The proposed method consists of three stages: candidate generation, scoring, and recipe ranking. At the candidate generation stage, a two-tower model using neural networks is used. This model allows for the effective integration of user and dish information into a common vector space and the quick retrieval of relevant recipes using a nearest neighbor search algorithm. After that, during the candidate generation stage, filtering takes place, which weeds out recipes that contain allergens or products restricted by the user. The scoring stage involves the user evaluating the recipes obtained from the candidate generation stage and forming a list of recommendations, which are ranked in order of recommendation in the next stage. For this purpose, machine learning models such as matrix factorization, decision tree-based models, and deep learning are used in the evaluation stage. These machine learning models take into account user preferences. The scoring stage also proposes modules that take into account contextual factors such as viewing history, day of the week, and time available for cooking. The ranking stage is implemented using a maximum marginal relevance algorithm, which ensures a balance between the relevance of recipe suggestions to user preferences and the diversity of recommendations. The scientific novelty of the work lies in the fact that, for the first time, a method has been developed that allows taking into account the user's personal preferences and their preferences regarding ingredients and cooking techniques. The proposed method has high potential for building intelligent personalized meal planning systems that promote a healthy lifestyle for users.

  • Open Access

    Model development of the system for avoiding echo chambers in social networks

    by Oleksii Stetsyk, Світлана Анатоліївна Теренчук, Олексій Андрійович Стецик

    This article examines the problems that appear in the process of the development of social networks. The focus of the study is on the echo chamber effect in various social networks. It turned out that the echo chamber effect is the result of the work of a deep neural network, which analyzes the interests and priorities of each social network user, checks other similar posts and priorities of other users, and then forms a “circle” of like-minded people for each user. Thus, the main drawback of the standard content generation algorithm is the selection of only publications and comments that support the user’s position. The goal of the study is a software architecture to solve the echo chamber problem in highly loaded social networks in real-time. The main idea was to cluster opinions on controversial topics and include user opinions from different clusters in user content. Data on controversial topics is collected from news, scholarly articles, publications, and comments with hashtags. The collected topics are clustered usi ng the K-means algorithm and the “elbow” method is used to find the optimal number of clusters. The result of the clustering of opinions by topic is provided as input data for the generation of a news feed by a recurrent neural network. The system uses Kafka as a message broker between microservices and AZURE blob storage for storing publications and comments. Both solutions support high scalability. У статті розглянуто можливості сучасних соціальних медіа та проблем, що з’являються в процесі розвитку соціальних мереж. Фокус дослідження спрямовано на ефект «бульбашки» в різних соціальних мережах. З’ясовано, що ефект «бульбашки» є результатом роботи глибинної нейронної мережі, яка аналізує інтереси і пріоритет кожного користувача соціальної мережі, перевіряє інші подібні публікації та пріоритети інших користувачів, після чого формує «коло» однодумців для кожного користувача. Таким чином в соціальних мережах з’являються кластери користувачів, які мають однакову думку щодо суперечливої теми. «Соціальні бульбашки» з’являються, коли кількість з'єднань між користувачами одного кластера величезна, а з’єднання між користувачами різних кластерів рідкісні. Отже, основним недоліком стандартного алгоритму генерації контенту є добірка лише публікацій і коментарів, що підтримують позицію користувача. Метою дослідження є архітектура програмного забезпечення для вирішення проблеми «бульбашки» у високонавантажених соціальних мережах у реальному часі. Основна ідея полягала в кластеризації думок у суперечливих темах і включення у вміст контенту користувача думок користувачів з різних кластерів. Дані щодо суперечливих тем збираються з новин, наукових статей, публікацій і коментарів із хеш тегами. Зібрані теми кластеризуються за допомогою алгоритму K-середніх. Для знаходження оптимальної кількості кластерів використовується метод «ліктя». Потім у різних кластерах користувачів розміщуються тексти з різними думками щодо певних тем. Результати кластеризації думок за темою надаються як вхідні дані для генерації стрічки новин рекурентною нейронною мережею. Отже, алгоритм генерації контенту вибрано як основний компонент вирішення проблеми виникнення «соціальної бульбашки». При цьому запропоновано удосконалити алгоритм генерації стрічки новин за допомогою машинного навчання і нейронних мереж. Також в цій статті запропоновано ефективну архітектуру системи для генерації стрічки новин. У системі використовується Kafka як посередник повідомлень між мікросервісами і Ажур блоб-сховище для зберігання публікацій і коментарів. Обидва рішення обґрунтовані високою масштабованістю. Реляційна база даних PostgreSQL використовується для метаданих публікацій, коментарів, користувачів та зв’язків між користувачами. При розробці архітектури системи передбачено виникнення певних проблем і окреслено шляхи їх вирішення. Наприклад, якщо в...


  • Open Access

    INTERNATIONAL COMPETITIVENESS AND RECOVERY STRATEGY OF THE AVIATION AND SPACE INDUSTRY OF UKRAINE

    by Olha Karpiak, Svitlana Terenchuk

    The influence of the aviation and space industries on the development of the world economy and economic development, industrial potential, competitiveness and position of various countries in the world is shown in the article. Global aviation and space market has been studied and an economic analysis of the aviation and space sector of the industry of the countries that are among ten world leaders in the industry was carried out. The main indicators of the development of the aviation and space industry of world countries-leaders and Ukraine have been systematized. The volumes of income of the leading companies of the industry producers, between which there is fierce competition, have been given. International competitiveness has been analyzed and ways of recovery and development of the aviation and space sector of Ukraine were proposed, taking into consideration the significant losses and damages caused by full-scale military invasion of aggressive neighbor on the territory of Ukraine. Scenarios of competitive behavior were considered and a strategy for the restoration of one of the largest enterprises in the world and single one in the aviation and space industry in Ukraine that has completed the cycle of engine production – Joint Stock Company Motor Sich has been made.