Petro Bidyuk
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Petro Bidyuk has a total of 62 co-authors. Here are the top co-authors by number of shared publications:
- Aleksandr Gozhyj
Petro Mohyla Black Sea National University
11 shared publications - Irina Kalinina
Petro Mohyla Black Sea National University
11 shared publications - Nataliia Kuznietsova
National Technical University of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute
11 shared publications - Yoshio Matsuki
National University of Kyiv-Mohyla Academy, Kyiv
9 shared publications - Oksana Tymoshchuk
National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute
6 shared publications - Oleksandr Terentiev
Institute of Telecommunications and Global Information Space of the National Academy of Sciences of Ukraine
6 shared publications
Latest Publications
- Open Access
- 10 November 2026
- Published by Publishing House Helvetica (Publications) in Information Technology: Computer Science, Software Engineering and Cyber Security
- No. 3,p. 3-14
ANALYSIS OF FINANCIAL RISKS INTERACTION
Financial risk management requires adequate modeling of multivariate distributions that reflect non-linear and asymmetric dependence structures, especially in the tails of distributions. Traditional correlation-based approaches are insufficient for capturing complex risk interactions. Therefore, copula theory provides a flexible framework for separating marginal behavior from dependence structure. The purpose of the work is to investigate the applicability of special copula functions for constructing formal multivariate probabilistic models of interacting financial risks and to evaluate their effectiveness in practical risk problems. The methodology is based on applying copula theory to construct multivariate probabilistic models of interacting financial risks. Joint distributions are modeled using generated three-dimensional Cauchy, Student-t, and normal distributions, as well as empirical currency exchange rate data. Heavy-tailed marginal distributions are built using threshold-based approaches and then combined into joint models through Archimedean (Gumbel, Clayton, Frank) and elliptical (Gaussian, Student-t) copulas. Parameter estimation is carried out using a two-step maximum likelihood procedure. The proposed framework is validated through computational experiments with both theoretical and empirical data. The scientific novelty consists in developing an integrated approach to modeling financial risk interaction by combining heavy-tailed marginal distributions with flexible copula-based dependence structures. The study refines the two-step maximum likelihood estimation procedure for joint models constructed from marginals and copulas and provides a comparative analysis of several copula families for capturing complex, non-linear, and asymmetric dependence. The results support the integration of the proposed approach into intellectual decision support systems for financial risk management. Conclusions. The study confirms the effectiveness of copula-based modeling for describing complex dependence structures between financial risks. The proposed combined marginal modeling and two-step parameter estimation procedure allow accurate construction of multivariate distributions, including tail dependence effects. The computational experiments demonstrate practical applicability of the approach. Future research will focus on expanding copula families and developing methods for estimating portfolio risk measures within an intellectual decision support framework.
- 10 November 2026
- Published by Publishing House Helvetica (Publications) in Information Technology: Computer Science, Software Engineering and Cyber Security
- No. 3,p. 3-14
Open AccessANALYSIS OF FINANCIAL RISKS INTERACTION
Financial risk management requires adequate modeling of multivariate distributions that reflect non-linear and asymmetric dependence structures, especially in the tails of distributions. Traditional correlation-based approaches are insufficient for capturing complex risk interactions. Therefore, copula theory provides a flexible framework for separating marginal behavior from dependence structure. The purpose of the work is to investigate the applicability of special copula functions for constructing formal multivariate probabilistic models of interacting financial risks and to evaluate their effectiveness in practical risk problems. The methodology is based on applying copula theory to construct multivariate probabilistic models of interacting financial risks. Joint distributions are modeled using generated three-dimensional Cauchy, Student-t, and normal distributions, as well as empirical currency exchange rate data. Heavy-tailed marginal distributions are built using threshold-based approaches and then combined into joint models through Archimedean (Gumbel, Clayton, Frank) and elliptical (Gaussian, Student-t) copulas. Parameter estimation is carried out using a two-step maximum likelihood procedure. The proposed framework is validated through computational experiments with both theoretical and empirical data. The scientific novelty consists in developing an integrated approach to modeling financial risk interaction by combining heavy-tailed marginal distributions with flexible copula-based dependence structures. The study refines the two-step maximum likelihood estimation procedure for joint models constructed from marginals and copulas and provides a comparative analysis of several copula families for capturing complex, non-linear, and asymmetric dependence. The results support the integration of the proposed approach into intellectual decision support systems for financial risk management. Conclusions. The study confirms the effectiveness of copula-based modeling for describing complex dependence structures between financial risks. The proposed combined marginal modeling and two-step parameter estimation procedure allow accurate construction of multivariate distributions, including tail dependence effects. The computational experiments demonstrate practical applicability of the approach. Future research will focus on expanding copula families and developing methods for estimating portfolio risk measures within an intellectual decision support framework.
- Open Access
- 30 September 2026
- Published by Igor Sikorsky Kyiv Polytechnic Institute in System research and information technologies
- No. 3,p. 94-105
Investigation of ensemble approaches and hybrid architectures for financial time series forecasting
The problem of short-term forecasting of financial time series characterized by nonlinearity, non-stationarity and sensitivity to external shocks is investigated. The application of ensemble methods of combining models (bagging, boosting, stacking), as well as the construction of hybrid neural architectures combining LSTM, GRU, CNN and Autoencoder are considered. A theoretical review of the conceptual features of these approaches is provided, their advantages and limitations are discussed, and the conditions for effective application are determined. A comparative analysis of the forecasting results using actual AMD data and four architectures is presented. It is shown that the hybrid Autoencoder + LSTM model achieves the highest accuracy, demonstrating the advantages of integrating mechanisms for automatic feature selection and modeling long-term dependencies. The obtained results confirm the feasibility of using hybrid ensemble approaches to improve the efficiency of financial forecasting.
- 30 September 2026
- Published by Igor Sikorsky Kyiv Polytechnic Institute in System research and information technologies
- No. 3,p. 94-105
Open AccessInvestigation of ensemble approaches and hybrid architectures for financial time series forecasting
The problem of short-term forecasting of financial time series characterized by nonlinearity, non-stationarity and sensitivity to external shocks is investigated. The application of ensemble methods of combining models (bagging, boosting, stacking), as well as the construction of hybrid neural architectures combining LSTM, GRU, CNN and Autoencoder are considered. A theoretical review of the conceptual features of these approaches is provided, their advantages and limitations are discussed, and the conditions for effective application are determined. A comparative analysis of the forecasting results using actual AMD data and four architectures is presented. It is shown that the hybrid Autoencoder + LSTM model achieves the highest accuracy, demonstrating the advantages of integrating mechanisms for automatic feature selection and modeling long-term dependencies. The obtained results confirm the feasibility of using hybrid ensemble approaches to improve the efficiency of financial forecasting.
- Open Access
- 29 September 2026
- Published by Igor Sikorsky Kyiv Polytechnic Institute in KPI Science News
- Vol. 144 (3)
AN APPROACH TO CONSTRUCTING AN INFORMATIVE FEATURE SPACE OF TECHNICAL INDICATORS FOR FINANCIAL PROCESSES BASED ON NEURAL NETWORK MODELS
Background. Financial time series are characterized by nonlinearity, non-stationarity, and a high level of noise, which significantly complicates their modeling and forecasting. The effectiveness of modern neural network models is largely determined not only by their architecture but also by the quality of the constructed input feature space. The use of technical indicators without accounting for their interdependencies may lead to multicollinearity, information redundancy, and a degradation of the models’ generalization ability. Objective. To develop an approach for constructing an informative feature space based on technical indicators and to investigate their impact on the forecasting accuracy of nonlinear and non-stationary financial processes using neural network models. Methods. The proposed approach is based on the principles of financial time series analysis and methods of data mining. For feature space construction, the approach utilizes an orthogonal combination of different indicator types — including trend, momentum, volatility, and relative price transformations. The investigation of their interaction mechanisms with the model was conducted using an LSTM architecture, focusing on the functional role of each parameter within the internal processes of information filtering and retention. Results. The developed methodology for constructing a feature space was tested on real-world stock price data for AMD using a hybrid architecture consisting of an LSTM-Autoencoder and an LSTM-Forecaster. The results demonstrated that forecasting performance strongly depends on the composition of the input feature vector. The best results were achieved by combining Price/MA and RSI, which increased the coefficient of determination from 0.8816 to 0.9316 and reduced the MAE from 3.1158 to 2.3569 compared with the baseline model. At the same time, the results confirmed that simply expanding the feature space does not guarantee improved forecasting performance due to potential information redundancy and interdependence among indicators. Conclusions. Improving the quality of forecasting financial processes can be achieved by constructing of informative and well-structured input feature space. The proposed approach of selective combination of technical indicators allows for reducing the impact of noise on model adequacy, increasing model stability, and improving its generalization ability. The obtained results create a foundation for practical implementation of high-performance automated decision support systems in financial markets.
- 29 September 2026
- Published by Igor Sikorsky Kyiv Polytechnic Institute in KPI Science News
- Vol. 144 (3)
Open AccessAN APPROACH TO CONSTRUCTING AN INFORMATIVE FEATURE SPACE OF TECHNICAL INDICATORS FOR FINANCIAL PROCESSES BASED ON NEURAL NETWORK MODELS
Background. Financial time series are characterized by nonlinearity, non-stationarity, and a high level of noise, which significantly complicates their modeling and forecasting. The effectiveness of modern neural network models is largely determined not only by their architecture but also by the quality of the constructed input feature space. The use of technical indicators without accounting for their interdependencies may lead to multicollinearity, information redundancy, and a degradation of the models’ generalization ability. Objective. To develop an approach for constructing an informative feature space based on technical indicators and to investigate their impact on the forecasting accuracy of nonlinear and non-stationary financial processes using neural network models. Methods. The proposed approach is based on the principles of financial time series analysis and methods of data mining. For feature space construction, the approach utilizes an orthogonal combination of different indicator types — including trend, momentum, volatility, and relative price transformations. The investigation of their interaction mechanisms with the model was conducted using an LSTM architecture, focusing on the functional role of each parameter within the internal processes of information filtering and retention. Results. The developed methodology for constructing a feature space was tested on real-world stock price data for AMD using a hybrid architecture consisting of an LSTM-Autoencoder and an LSTM-Forecaster. The results demonstrated that forecasting performance strongly depends on the composition of the input feature vector. The best results were achieved by combining Price/MA and RSI, which increased the coefficient of determination from 0.8816 to 0.9316 and reduced the MAE from 3.1158 to 2.3569 compared with the baseline model. At the same time, the results confirmed that simply expanding the feature space does not guarantee improved forecasting performance due to potential information redundancy and interdependence among indicators. Conclusions. Improving the quality of forecasting financial processes can be achieved by constructing of informative and well-structured input feature space. The proposed approach of selective combination of technical indicators allows for reducing the impact of noise on model adequacy, increasing model stability, and improving its generalization ability. The obtained results create a foundation for practical implementation of high-performance automated decision support systems in financial markets.
- Open Access
- 1 May 2026
- Published by Kyiv National University of Construction and Architecture in Environmental safety and natural resources
- Vol. 58 (2) , 117-134
Financial risk analysis in multidimensional systems
A new approach is proposed for modeling the interdependence among factors of multivariate risks, represented as matrices of interdependence measures for numerical description and a family of copulas with parameter estimates for analytical description. The approach proposed to construct a multivariate risk model in which, marginal distributions are modeled separately using elliptical distributions for measurements at the center of the samples and extreme distributions in the tails, while the dependencies between risks are modeled by copulas. The joint distribution is modeled using marginal distributions and copulas and can be applied to the analysis of risk characteristics. An approach to determining risk dependencies using the concept of mutual information within the framework of Bayesian networks has been developed. A computational experiment involving two generated, theoretically well-known three-dimensional distributions and one empirical three-dimensional distribution for exchange rates demonstrated the applicability of the proposed approach to modeling multidimensional risk.The problem of identifying the optimal portfolio structure under active risk management and asset liquidity constraints, a multidimensional model for estimating tail risk measures is proposed. A computational experiment conducted to estimate risk measures by generating a sample yielded an estimation error of less than one percent for non-extreme quantiles. The quality of the estimation of risk deviation measures requires further refinement of the model. The quality of risk measure estimates for the tail regions of distributions indicates that the model based on a combination of marginal distributions using normal and Pareto distributions needs to be improved to describe central observations.
- 1 May 2026
- Published by Kyiv National University of Construction and Architecture in Environmental safety and natural resources
- Vol. 58 (2) , 117-134
Open AccessFinancial risk analysis in multidimensional systems
A new approach is proposed for modeling the interdependence among factors of multivariate risks, represented as matrices of interdependence measures for numerical description and a family of copulas with parameter estimates for analytical description. The approach proposed to construct a multivariate risk model in which, marginal distributions are modeled separately using elliptical distributions for measurements at the center of the samples and extreme distributions in the tails, while the dependencies between risks are modeled by copulas. The joint distribution is modeled using marginal distributions and copulas and can be applied to the analysis of risk characteristics. An approach to determining risk dependencies using the concept of mutual information within the framework of Bayesian networks has been developed. A computational experiment involving two generated, theoretically well-known three-dimensional distributions and one empirical three-dimensional distribution for exchange rates demonstrated the applicability of the proposed approach to modeling multidimensional risk.The problem of identifying the optimal portfolio structure under active risk management and asset liquidity constraints, a multidimensional model for estimating tail risk measures is proposed. A computational experiment conducted to estimate risk measures by generating a sample yielded an estimation error of less than one percent for non-extreme quantiles. The quality of the estimation of risk deviation measures requires further refinement of the model. The quality of risk measure estimates for the tail regions of distributions indicates that the model based on a combination of marginal distributions using normal and Pareto distributions needs to be improved to describe central observations.
- Open Access
- 29 December 2025
- Published by Igor Sikorsky Kyiv Polytechnic Institute in System research and information technologies
- No. 4,p. 58-70
Analysis of actuarial risk with generalized linear models
The problem of applying generalized linear models to the analysis of actuarial risks in the context of premium charges to clients was considered. The Monte-Carlo method for Markov chains was applied. Two situations were considered for the computational experiment. For the first one, insurance indicators and the target variable were randomly assigned due to the problem of public data access. To create three datasets, charges were generated from normal, gamma, and Pareto distributions with dynamic variance, and noise was added to stimulate a non-stationary process. In the second situation, actual actuarial data from the Singa-pore Actuarial Society was used. Generalized Linear Models with normal dis-tribution and logarithmic link function, an exponential distribution and loga-rithmic link function, and Laplace distribution with identity link function were constructed. Based on the model-fitting quality metrics, conclusions were drawn about their structure.
- 29 December 2025
- Published by Igor Sikorsky Kyiv Polytechnic Institute in System research and information technologies
- No. 4,p. 58-70
Open AccessAnalysis of actuarial risk with generalized linear models
The problem of applying generalized linear models to the analysis of actuarial risks in the context of premium charges to clients was considered. The Monte-Carlo method for Markov chains was applied. Two situations were considered for the computational experiment. For the first one, insurance indicators and the target variable were randomly assigned due to the problem of public data access. To create three datasets, charges were generated from normal, gamma, and Pareto distributions with dynamic variance, and noise was added to stimulate a non-stationary process. In the second situation, actual actuarial data from the Singa-pore Actuarial Society was used. Generalized Linear Models with normal dis-tribution and logarithmic link function, an exponential distribution and loga-rithmic link function, and Laplace distribution with identity link function were constructed. Based on the model-fitting quality metrics, conclusions were drawn about their structure.
- 12 November 2025
- Published by Springer Nature
- p. 71-87
A Systematic Approach to Modeling and Forecasting Based on Real Data in Machine Learning Tasks
- 12 November 2025
- Published by Springer Nature
- p. 71-87
A Systematic Approach to Modeling and Forecasting Based on Real Data in Machine Learning Tasks
- 21 October 2025
- Published by Academy of Cognitive and Natural Sciences
- p. 125-134
Survival Models as Copulas for Green Risks Prediction
This work provides a comprehensive overview of the current state of copulas and proposes to use survival models for newly introduced families. While the copula’s appearance has seen widespread application in various fields, including environmental sciences (particularly hydrology), economics, and as enhancements to existing models to improve performance. The extensive range of potential copulas, which can be systematically expanded through rigorously defined procedures such as those for the Archimedean class, underscores their versatility. However, many existing approaches emphasize forms of copulas that focus on past values, providing insights into the likelihood of events occurring before the value of interest. This paper introduces an alternative approach aimed at assessing the likelihood of events occurring in the future. This forward-looking perspective leverages the strengths of copulas more effectively, particularly in risk- sensitive fields such as environmental management. To accomplish it, the newly developed copula is shown to be transformed in a survival form.
- 21 October 2025
- Published by Academy of Cognitive and Natural Sciences
- p. 125-134
Survival Models as Copulas for Green Risks Prediction
This work provides a comprehensive overview of the current state of copulas and proposes to use survival models for newly introduced families. While the copula’s appearance has seen widespread application in various fields, including environmental sciences (particularly hydrology), economics, and as enhancements to existing models to improve performance. The extensive range of potential copulas, which can be systematically expanded through rigorously defined procedures such as those for the Archimedean class, underscores their versatility. However, many existing approaches emphasize forms of copulas that focus on past values, providing insights into the likelihood of events occurring before the value of interest. This paper introduces an alternative approach aimed at assessing the likelihood of events occurring in the future. This forward-looking perspective leverages the strengths of copulas more effectively, particularly in risk- sensitive fields such as environmental management. To accomplish it, the newly developed copula is shown to be transformed in a survival form.
- Open Access
- 30 September 2025
- Published by Kyiv National University of Construction and Architecture in Environmental safety and natural resources
- Vol. 55 (3) , 33-49
Mathematical modeling, intelligent data analysis and artificial intelligence to support decision-making in post-war reconstruction
The article is devoted to the actual scientific and applied problem – development of information technologies of support of making management decisions in conditions of systemic uncertainty, in particular, characteristic for periods of martial law and post-war recovery. Its peculiarity is complexity of formalization and structuring, necessity of solution of tasks of decision-making in conditions of systemic uncertainty and risk caused by military aggression. Therefore, the actual problem is development of scientifically based flexible and universal methodology of application of mathematical models, methods of intellectual data analysis, artificial intelligence, information technologies for formation of management decisions, within the framework of the corresponding decision support system (DSS). In the work, the architecture of the corresponding DSS is proposed, the key link of which is subsystem of preliminary processing of large volumes of input data formed from statistical indicators, results of observations and surveys, information from Internet sources, etc., for construction of models of the studied processes, development of scenarios, strategic and operational planning and forecasting. The system implements all elements of the technological chain of collecting and processing structured and unstructured data, mathematical modeling, methods of intelligent data analysis, scenario analysis, cognitive modeling, artificial intelligence, etc. A feature of the developed methodology is the ability to adequately assess the current situation, its retrospective and predict its development and consequences under several scenarios, which is relevant for decision-making tasks in martial law and post-war recovery. The proposed information technologies are intended for use in state and public administration systems. Their structure is flexible and adaptive: the components can be used separately or as part of existing decision-making support systems, implementation will increase the quality and efficiency of management decisions by optimizing analytical processes and data processing speed.
- 30 September 2025
- Published by Kyiv National University of Construction and Architecture in Environmental safety and natural resources
- Vol. 55 (3) , 33-49
Open AccessMathematical modeling, intelligent data analysis and artificial intelligence to support decision-making in post-war reconstruction
The article is devoted to the actual scientific and applied problem – development of information technologies of support of making management decisions in conditions of systemic uncertainty, in particular, characteristic for periods of martial law and post-war recovery. Its peculiarity is complexity of formalization and structuring, necessity of solution of tasks of decision-making in conditions of systemic uncertainty and risk caused by military aggression. Therefore, the actual problem is development of scientifically based flexible and universal methodology of application of mathematical models, methods of intellectual data analysis, artificial intelligence, information technologies for formation of management decisions, within the framework of the corresponding decision support system (DSS). In the work, the architecture of the corresponding DSS is proposed, the key link of which is subsystem of preliminary processing of large volumes of input data formed from statistical indicators, results of observations and surveys, information from Internet sources, etc., for construction of models of the studied processes, development of scenarios, strategic and operational planning and forecasting. The system implements all elements of the technological chain of collecting and processing structured and unstructured data, mathematical modeling, methods of intelligent data analysis, scenario analysis, cognitive modeling, artificial intelligence, etc. A feature of the developed methodology is the ability to adequately assess the current situation, its retrospective and predict its development and consequences under several scenarios, which is relevant for decision-making tasks in martial law and post-war recovery. The proposed information technologies are intended for use in state and public administration systems. Their structure is flexible and adaptive: the components can be used separately or as part of existing decision-making support systems, implementation will increase the quality and efficiency of management decisions by optimizing analytical processes and data processing speed.
- 26 April 2025
- Published by Springer Nature
- p. 206-219
A Systematic Approach to Data Normalization and Standardization in Machine Learning Problems
- 26 April 2025
- Published by Springer Nature
- p. 206-219
A Systematic Approach to Data Normalization and Standardization in Machine Learning Problems