KPI Science News

DYNAMIC LOGISTIC REGRESSION MODELING FOR BANKRUPTCY RISK PREDICTION IN UKRAINIAN BUILDING SECTOR

Vadiv Romanuke — Mon, 30 Mar 2026 00:00:00 +0300

Background. Financial distress and bankruptcy forecasting have become increasingly important in the context of post-war economic recovery and restructuring of Ukrainian industries. Firms in the building-and-construction materials sector operate under high uncertainty, where early detection of insolvency risk is crucial for maintaining financial stability. Logistic regression models, widely used in environmental and risk analytics, can be adapted to represent the nonlinear transition from solvency to bankruptcy as a probabilistic process.

Objective. The objective is to develop and evaluate both static and dynamic logistic regression models for predicting potential bankruptcy of a representative Ukrainian building-materials manufacturer. The dynamic extension is aimed at capturing temporal persistence in financial performance through lagged predictors.

Methods. A synthetic monthly dataset (5 years, 60 observations) is generated to simulate realistic financial ratios, including liquidity, leverage, profitability, efficiency, and interest coverage (solvency). The models are estimated in MATLAB using maximum-likelihood logistic regression with L2 regularization (ridge penalty) to retain correlated predictors. The dynamic model incorporated one-period lags of all financial ratios and the one-period-lagged response. Predictive performance is assessed by accuracy, precision, recall, F1-score, and the confusion matrix.

Results. The static logistic model achieved an average accuracy of approximately 89 %, but it missed two bankruptcy-risky months out of six ones. The dynamic model improved performance to 94 % accuracy, without missing a bankruptcy-risky month, but falsely labeling a non-risky month as bankruptcy-risky one. The signs of estimated coefficients are consistent with economic logic: higher leverage increases bankruptcy probability, whereas greater liquidity, profitability, efficiency, and solvency reduce it.

Conclusions. Dynamic L2-regularized logistic regression provides an interpretable and computationally efficient framework for early bankruptcy prediction in Ukrainian industrial firms. The inclusion of lagged financial indicators enhances predictive stability and timeliness, enabling practical early-warning applications.

METHOD FOR DETECTING CORNER POINTS IN IMAGES USING A COROTATIONAL BEAM SPLINE

Igor Orynyak, Dmytro Koltsov, Danylo Tavrov — Mon, 30 Mar 2026 00:00:00 +0300

Background. The detection of corner points in images is of great importance for object identification and has numerous applications in computer vision and pattern recognition. Typically, this task is performed using geometric analysis of black-and-white contours, to which Gaussian smoothing is subsequently applied in order to identify points of maximum curvature. Functional minimization is then applied to these regions to determine the angle magnitude between adjacent contour segments. A drawback of this approach lies in the difficulty of accounting for artifacts, as well as in the fact that increased smoothing leads to a reduction in the effective scale (size) of the image.

Objective. To develop a method for detecting corner points in a digitized grayscale image using the corotational beam spline (CBS) method.

Methods. Application of the CBS method using the proposed corner dummy points, for which the angular continuity condition is not satisfied; instead, zero curvature is postulated, resulting in a discontinuity (jump) in the tangent direction at that point.

Results. The CBS method is applied to smooth the contour of a black-and-white image according to the length of a contour segment rather than the number of points on that segment, thereby preserving the overall image scale. Special corner dummy points are proposed that allow for a loss of angular continuity. Candidates for corner points are identified based on the local extrema of the curvature graph. For each point, the work is defined as the square of the distance between the point and its corresponding location on the contour, multiplied by the length of the segment associated with that point. The concept of integral work is introduced as the sum of individual works within the region of maximum curvature. A criterion for the existence of a corner point is developed based on the analysis of the ratio of individual works obtained in the absence and in the presence of a corner dummy point.

Conclusions. The application of adaptive smoothing according to the distance between points on the contour, which are projections (correspondences) of the measured points, enables the method to be applied to datasets with varying point densities, thereby improving the quality of the reconstructed contour. The use of corner dummy points that allow for a loss of angular continuity makes it possible to more accurately reproduce the target contour, in particular sharp angle changes. Furthermore, the use of the ratio of individual works obtained in the absence and in the presence of a dummy point serves as a reliable criterion for corner point detection. The ratio of work values demonstrates an improvement by a factor of 8–30 for corner points, whereas for curved regions this value deteriorates or remains nearly unchanged.

A METHOD FOR ANALYZING MULTICOMPONENT COMPLEX SYSTEMS BASED ON MULTIDIMENSIONAL MATRICES OF ADJACENCY AND ENTROPY RELATIONS BETWEEN SYSTEM COMPONENTS

Vyacheslav Voloshyn, Illia Tkalenko — Mon, 30 Mar 2026 00:00:00 +0300

Background. Existing methods for analysing multicomponent complex systems account only weakly for the relative incompatibility of system parameters, in particular the interplay between physical, chemical, and other mechanisms and the configurational, thermodynamic, and additional properties of the system. The challenge of achieving a stable, practically formalised investigation of the development of complex systems—through a mathematically simplified and conceptually transparent set of indicators enabling comprehensive analysis—remains unresolved.

Objective. The development of an alternative method for maximally comprehensive and simplified analysis of multicomponent complex systems, based on their functionality, structural organisation, and inherent properties.

Methods. A method for analysing complex systems by means of multidimensional entropy-structured adjacency matrices is proposed as a derivative model for describing their multicomponent nature. Entropy and its variation, which reflects changes in the state of the system itself, are proposed as thermodynamic indicators of multicomponent systems.

Results. The first essential feature of the model lies in the partitioning of the total entropy of the entire system into superpositional components corresponding to the parameters of the system’s properties, which forms the basis for a certain universality of the method. The second significant distinction of the model is associated with its multilayered nature within the adjacency matrix. This enables the investigation of a wide spectrum of incomparable properties of a complex system, including its structural organisation, informational characteristics, functionality, physico-chemical features, surface phenomena—including the boundary with the supersystem—as well as its capacity for thermodynamic nonequilibrium and kinetic behaviour, depending on the nature of the system itself. The validation and advantages of the model are demonstrated through several practical examples from the field of metallurgical waste recycling, where mixtures are considered as multicomponent systems used in the production of metallurgical briquettes as secondary raw materials for this industrial sector.

Conclusion. A qualitatively new approach has been developed for the study of multicomponent complex systems by means of a multidimensional entropy-containing adjacency matrix, which describes interrelations through the total entropy decomposed into individual components. Its relative universality and practical applicability are demonstrated.

INTELLIGENT SYSTEM FOR ADAPTIVE CONTROL OF THE TECHNOLOGICAL PROCESS OF LAYING ASPHALT CONCRETE BASED ON NEURAL NETWORK MODELS

Yaroslav Steshenko, Anatolii Protasov — Mon, 30 Mar 2026 00:00:00 +0300

Background. The quality and durability of asphalt concrete pavement directly depends on compliance with technological parameters during its laying process. Any violation of such parameters inevitably leads to the appearance of hidden defects. Traditional automated technological process control systems are often based on rigid algorithms that are unable to fully take into account the complex nonlinear effects of air humidity, material cooling rate and dynamic base stiffness. The implementation of closed-loop control loops based on feedback ensures the stability of the laying parameters regardless of external disturbances. However, the effectiveness of such systems is significantly limited without the integration of predictive models that can detect potentially defective areas even before the final cooling of the mixture.

Objective. The purpose of the study is to develop and analyze the effectiveness of an intelligent adaptive control system (IACS) for the asphalt concrete paving process, which integrates a predictive neural network model into a closed-loop control loop.

Methods. To implement the system, object-oriented programming methods (Python language) and machine learning libraries (XGBoost, TensorFlow) were used. The methodology is based on comparative computer modeling of 1000 technological cycles. The results were verified by comparing the predicted density values with reference physical and mathematical models of compaction.

Results. The implementation of the proposed system allowed to increase the coating density by 4 times compared to standard systems. The reduction of the reaction time to temperature disturbances from 18.5s to 5.2s confirms the ability of the system to act proactively and reduce the probability of hidden damage to the level of 1.8%. The value of the process stability coefficient.

Conclusions. The proposed intelligent adaptive control system allows solving the problem of delayed response of traditional automated complexes to stochastic changes in external factors during asphalt concrete laying. The proposed approach eliminates the negative impact of temperature instability and humidity fluctuations, which usually lead to the appearance of hidden defects and heterogeneity of the coating structure. Thanks to the integration of predictive neural network models into a closed control loop, it was possible to provide proactive process control, where parameter adjustment occurs based on the forecast of the material state, and not only on the fact of deviation from the norm.

VARIO LENSES AS FIXED ELEMENTS OF ZOOM-OPTICAL SYSTEMS

Igor G. Chyzh — Mon, 30 Mar 2026 00:00:00 +0300

Baсkground. The modernisation and structural improvement of zoom optical systems by utilising fixed vario lenses that are either electrically or mechanically controlled.

Objective. An evaluation of the accomplishments in the advancement and enhancement of adaptive vario lenses featuring externally regulated optical power, intended for integration into zoom optical systems as permanent elements. An analysis of the factors that could result in the instability of the cardinal points within the optical system of vario lenses.

Methods. An examination of the physical principles governing contemporary externally controlled vario lenses, with a focus on the conditions that facilitate alterations in their optical power. Application of the fundamental tenets of optical system theory to evaluate how variations in the geometric configuration and placement of the interface between the operational optical media of vario lenses influence their optical power. A thorough analysis and organization of information available on manufacturers' websites, as well as in public information sources and publications from varifocal lens developers, regarding the design characteristics of their existing models.

Results. A comprehensive analysis of the mechanisms responsible for alterations in the optical power of membrane and electrically wetted vario lenses revealed that fluctuations in their optical power result from modifications in the radius of curvature of the surface separating the working optical media, as well as by changes in the position of this surface along the optical axis. This leads to a shift along the optical axis of the vario lens of the cardinal (principal and nodal) points of the vario lens and a corresponding shift of their focal points.

Conclusions. In the examined membrane and electrowetting vario lenses, alterations in the geometric shapes of the space occupied by the working optical media affects not only the optical power, but also the position of the cardinal points of the vario lenses. Therefore, in zoom-optical systems, vario lenses cannot be regarded as fixed optical components despite the fact that their physical locations within the system design are actually fixed. The shift factor of the cardinal points of the vario lenses requires comprehensive further investigation and determination of the functional dependence between the specified shift and the values of the optical power of the vario lense. The practical application of the findings from these investigations will allow improving the algorithm of parametric synthesis of zoom-optical systems with vario lenses built into them, and will also provide a theoretical foundation for more accurate programming of the level of control electrical signals supplied from microcontrollers to the vario lenses. This will aid in averting detrimental axial shifts of the images formed by them (image defocusing) and disruption of the aberration correction of the system during the operation of zoom optical systems

ENGINEERING OF SOFTWARE SYSTEMS BASED ON SNAPSHOT-CENTRIC CQRS WITH EVENT SOURCING ARCHITECTURE

Dmytro Hruzin, Oleksandr Lytvynov — Mon, 30 Mar 2026 00:00:00 +0300

Background. Command Query Responsibility Segregation (CQRS) with Event Sourcing (ES) is a widely adopted solution for designing scalable and high-performance information systems. However, classical CQRS with ES implementations are often associated with increased complexity in development and maintenance.

Objective. The goal of this study is to simplify the development and maintenance of software systems built on CQRS with ES by introducing an alternative variation of the architecture.

Methods. The classical architectural variation was analyzed, and the components that increase the complexity of system development and maintenance were identified. Based on this analysis, an alternative architectural variation (mCQRS) is proposed, that uses a lower-complexity component set. The solution is based on a relational database in which aggregate state snapshots are treated as the source of truth, thereby reducing implementation and maintenance complexity and facilitating potential migration to other architectural variations.

Results. Representative test projects were developed for both the classical and the proposed CQRS with ES variations. Cyclomatic complexity values for a typical command execution workflow (120 for Classical CQRS and 82 for mCQRS) indicate a 31.67% decrease in complexity. Performance measurements show that server response time for queries is identical for both variations (44 ms), whereas the end-to-end time to reach system consistency for commands is 268 ms for Classical CQRS and 347 ms for mCQRS, corresponding to a 22.76% decrease in performance. Despite this degradation, write-operation throughput remains high in the context of established industry practices.

Conclusions. The proposed approach improves the efficiency of development and maintenance and reduces the required level of developer expertise; it is suitable for systems in which write-operation performance is not critical