System research and information technologies

Method of polarization selection of navigation objects in adverse weather conditions using statistical properties of radio signals

Fri, 28 Mar 2025 00:00:00 +0200

This research article is devoted to studying and applying polarization selection for navigation objects in difficult atmospheric conditions. It provides a novel application of Stokes parameters in radar signal processing for navigation objects, validated by experimental data. The main emphasis is on using the statistical properties of the polarization parameters of partially polarized echo signals. The article discusses in detail the statistical properties of the polarization parameters of partially polarized echo signals, which can be used to improve the accuracy of ship radiolocation systems. The study is based on analyzing experimental data collected in various atmospheric conditions. The results indicate the effectiveness of polarization selection in improving the stability and accuracy of radar navigation systems in various atmospheric conditions. The use of statistical methods allows the navigation system to adapt to changing conditions, ensuring reliability in different scenarios. Polarization selection based on the statistical properties of polarization parameters is a promising method to improve navigation in high atmospheric humidity, fog, and other complex atmospheric conditions. It can be used in the development of modern navigation systems.

Optimal selection of cotton warp sizing parameters under system research limitation

Hanna Tkachuk, Vadim Romanuke, Andriy Tkachuk — Fri, 28 Mar 2025 00:00:00 +0200

Warp sizing is the process of applying the sizing agents to the warp yarn to improve its weavability along with improving the economic performance of weaving. We consider a finite set of sizing agents or parameters mapped into a finite set of sizing quality indicators. Due to various limitations of material and time resources, exhaustive system research and constructing an information technology to interpret and optimize sizing data is impossible. Therefore, we suggest an algorithm for controlling warp sizing quality under system research limitation, where optimal selection of cotton warp sizing parameters is exemplified. The algorithm utilizes a set of basis vectors of sizing parameters corresponding to a set of respective vectors of quality indicators. The method of radial basis functions is used to determine the probabilistically appropriate vector of quality indicators for any given vector of sizing parameters. The uncountably infinite space of sizing vectors is uniformly sampled into a finite space. The finite space may be refined by excluding sizing vectors corresponding to inadmissible values of one or more quality indicators. A set of Pareto-efficient sizing vectors is determined within the finite (refined) space, and an optimal, efficient sizing vector is determined as one being the closest to the unachievable sizing vector. The suggested algorithm serves as a method of optimal selection of warp sizing parameters, resulting in improved performance of warp yarns that can withstand repeated friction, stretching, and bending on the loom without causing a lot of fluffing or breaking. The algorithm is not limited to cotton, and it can be applied to any yarn material by an experimentally adjusted radial basis function spread.

Efficiency comparison of missing data imputation methods in predictive model creation

Andrii Popov — Fri, 28 Mar 2025 00:00:00 +0200

Missing data is a common issue in data analysis and machine learning. This article analyzes the impact of missing data imputation methods during the data preprocessing stage on the quality of forecasting models. Selected methods are listwise deletion, mean imputation, and two implementations of the multiple imputation method in Python and R languages. Selected classifiers are Logistic Regression, Random Forest, Support Vector Machine, and Light Gradient Boosting Machine. The performance quality of forecasting models is estimated using accuracy, precision, and recall metrics. Two datasets were used as binary classification problems with different target metrics. The highest performance was achieved when the R implementation of the multiple imputation method was combined with RF and LGBM classifiers.

Identification of nonlinear systems with periodic external actions (Part III)

Viktor Gorodetskyi — Fri, 28 Mar 2025 00:00:00 +0200

The article considers the problem of identifying a mathematical model in the form of a system of ordinary differential equations. The identified system can have constant and periodic coefficients. The source of information for solving the problem is time series of observed variables. The article studies the effect of uniformly distributed noise on the identification result. To solve the problem, the algorithm proposed by the author in previous works was used. It is shown that the method has different sensitivity to noise depending on which of the observed variables is contaminated with noise. The implementation of the method is illustrated by numerical examples of identifying nonlinear differential equations with polynomial right-hand sides.

Numerical algorithm for calculation of the vacuum conductivity of a non-linear channel for transporting a short-focus electron beam in the technological equipment

Igor Melnyk, Alina Pochynok, Mykhailo Skrypka — Fri, 28 Mar 2025 00:00:00 +0200

In the article, based on solving the equations of vacuum technology, an iterative algorithm for calculating vacuum conductivity and the geometric parameters of a curvilinear channel for transporting a short-focus electron beam is proposed and studied. For such a type of channel, the dependence of its radius on the longitudinal coordinate is described by a power function. The proposed algorithm is based on the numerical solution of a set of nonlinear equations using the Steffensen method. The results of the test calculations are presented. The provided tests confirm the stability of the proposed algorithm’s convergence for correct pressure and pumping speed values in electron-beam technological equipment. Such curved transport channels can be used in electron beam equipment based on high-voltage glow discharge electron guns intended for welding, melting metals, and the deposition of thin films. The criterion for the optimal geometry of a nonlinear channel is the minimum power loss of the electron beam during its transportation while ensuring the required pressure drop between the discharge and technological chambers.

Agent-based approach to implementing artificial intelligence (AI) in service-oriented architecture (SOA)

Anatolii Petrenko — Fri, 28 Mar 2025 00:00:00 +0200

Artificial Intelligence (AI) is becoming a general-purpose technology and is gaining a universal character for engineering, science, and society that today is only inherent in mathematics and computer technology. The agent-based approach to implementing artificial intelligence (AI) within the service-oriented architecture of an application is a fascinating and highly synergistic concept. Combining these paradigms leads to robust, scalable, and intelligent systems well suited for dynamic and distributed environments. This paper presents the results of a comparative analysis of three possible approaches to integrating AI into business processes, namely, connecting AI agents to service-oriented architecture (SOA), connecting AI agents to software (SaaS), and building AI as a service (AIaaS). The paper provides some insights into the potential benefits, challenges, examples, and considerations when adopting each of these approaches.

Investigation of the effectiveness of artificial neural networks of different generations in the task of forecasting in the financial sphere

Yevgeniy Bodyanskiy, Yuriy Zaychenko, Helen Zaichenko, Oleksii Kuzmenko — Fri, 28 Mar 2025 00:00:00 +0200

This paper discusses ANNs of different generations. The efficiency of using computational intelligence in the task of short- and medium-term forecasting in the financial sphere is investigated. For the investigation, a fully connected feed-forward network (Back Propagation), a recurrent network (LSTM), a hybrid deep learning network based on self-organization (GMDH neo fuzzy), and a hybrid system of computational intelligence based on bagging and group method of data handling (HSCI bagging) were chosen. The experimental parameters chosen are the prediction interval, the number of inputs, the percentage of validation data in the training set, and the number of fuzzifiers (for GMDH neo-fuzzy). Experiments were conducted, and the best results for different prediction intervals were compared. The optimal parameters of the networks and the feasibility of their use in the task of forecasting at different intervals are determined.

Assessing the impact of AI-generated product names on e-commerce performance

Oleksandr Bratus — Fri, 28 Mar 2025 00:00:00 +0200

This paper studies the impact of Large Language Model (LLM) technology on the e-commerce industry. This work conducts a detailed review of the current implementation level of LLM technologies in the e-commerce industry. Next, it analyzes the approaches to detecting AI-generated text and determines the limitations of their application. The proposed methodology defines the impact of LLM models on the e-commerce industry based on a comparative analysis between indicators of machine-generated texts and e-commerce product metrics. Applying this methodology to real data, one of the most relevant data collected after the release of ChatGPT, the results of statistical analyses show a positive correlation between the studied indicators. It is proved that this dependence is dynamic and changes over time. The obtained implicit indicators measure the influence of LLM technologies on the e-commerce domain. This influence is expected to grow, requiring further research.

The hybrid sequential recommender system synthesis method based on attention mechanism with automatic knowledge graph construction

Dmytro Androsov, Nadezhda Nedashkovskaya — Fri, 28 Mar 2025 00:00:00 +0200

Sequential personalized recommendations, such as next best offer prediction or modeling demand evolution for next basket prediction, remain a key challenge for businesses. In recent years, deep learning models have been applied to solve these problems and demonstrated high feasibility. With the introduction of graph-based deep learning, it has become easier to perform collaborative filtering and link prediction tasks. The current paper proposes a new method of building a recommender system using a graph representation learning framework in combination with deep neural networks for sequence-to-sequence modeling and statistical learning for sequence-to-graph mapping. Benchmarking model performance on an online retail store visits dataset provides evidence of the method’s ranking capabilities.

Studying the relationship between tuberculosis and socioeconomic, medical, and demographic factors in Ukraine

Denys Nevinskyi, Dmytro Martjanov, Ihor Semianiv, Yaroslav Vyklyuk — Fri, 28 Mar 2025 00:00:00 +0200

Ukraine is currently experiencing a new, ongoing tuberculosis offensive. Our study analyzes the impact of various socioeconomic and medical factors, including the number of specialized hospitals, fluoroscopic examinations of the population, the number of healthcare workers, the level of alcohol and drug abuse, and others, on the prevalence of tuberculosis among different demographic groups in Ukraine. Artificial intelligence methods made it possible to identify key factors contributing to the growth or decline in tuberculosis incidence. The results of the SHAP (SHapley Additive exPlanations) analysis, which offers a methodology for interpreting complex machine learning models, shows the most important factors that influence the incidence of tuberculosis in Ukraine. The sensitivity analysis provided more important and detailed information, which confirmed the results of the SHAP analysis.