Information and Software Technologies

Feature standardization is a widely applied step in data preprocessing that transform the feature values so that the data mean value is equal to zero and standard deviation is equal to one. It allows for easier and more comprehensible comparison between features by making each feature contribute evenly to the process of analysis. The aim of this paper is to investigate the impact of data standardization on the performance of principal component analysis (PCA) and classification using support vector machine (SVM) algorithm. The analysis is performed on two datasets with overlapping feature values and more separately spaced feature values for unstandardized and standardized data. Additionally, balanced and imbalanced datasets are evaluated. Final results showcase that data standardization significantly improves the performance of PCA as well as classification effectiveness of the SVM algorithm.

Effective route planning, despite numerous studies, is still a great challenge. Given real-world data and demanding criteria, the problem can be even more complex, with no single optimal solution. In this paper, an interactive system with dedicated Dijkstra-based algorithm is proposed to tackle the problem of efficient bicycle route planning. With the presented web application framework, the user is able to generate a complete cycle of the desired length and surface quality, complementing the selected profile of the bicycle. The results indicate the reliability of the system. The routes adapt well to the chosen profile and fall within the provided acceptable total-length error (5%), reaching 3.24 MAPE for shorter and 2.74 MAPE for longer routes tested.

Classification of aquatic creatures is essential for biodiversity research in ecosystems. The minor differences between species and the underwater conditions affecting image quality pose significant challenges for image classification models. This type of classification, where images belong to a single metaclass, is referred to as Fine-Grained Visual Classification (FGVC). A key aspect of FGVC is the ability to distinguish these subtle differences that determine membership in a specific class. Therefore, in this study, an attention-based CNN model with a skip-connection mechanism was proposed. The model was tested on seven publicly available datasets, and for four of them - Croatian Fish Dataset, DeepFish, Large-Scale Fish, and Fish-Pak - it achieved the highest score, respectively 96.88%, 100%, 100%, and 100%. The presented model is significantly lighter compared to existing methods: it has 1.4 million parameters, which, combined with very high accuracy, enables its application in resource-constrained environments.

The motor with permanent magnets used for field excitation, including the BLDC motor, currently finds numerous applications, particularly in electric vehicle propulsion. These motors are powered by an electrical system called an electronic commutator. It is justified to investigate the efficiency of converting electrical energy into mechanical energy depending on the method of supplying the motor windings. Speed regulation requires changing the average voltage supplied to the motor windings. The average voltage supplied to the motor can be changed in two ways: by pulse width modulation (PWM) implemented on the transistors of the electronic commutator or by using an additional DC/DC converter connected between the power source and the electronic commutator. The article presents a research setup examining the total energy losses in the studied BLDC motor depending on the method of power supply, load torque, and rotational speed.

In the management of kidney stones, identifying the stones’ composition is crucial for determining appropriate treatment methods, as different compositions may require different treatment strategies. This study introduces a novel approach utilising Fourier Transform Infrared (FTIR) spectra data to determine kidney stones’ composition. By using non-negative matrix factorisation (NMF) for dimensionality reduction and Gaussian mixture model (GMM) for Bayesian soft classification, this method aims to offer a faster and non-destructive alternative to other traditional methods, such as wet chemistry. Initial findings indicate that this approach can effectively analyse kidney stones’ composition, showing promise even in cases with limited sample sizes. The application of NMF and GMM to FTIR data could significantly improve the efficiency and accuracy of kidney stones’ composition analysis, potentially leading to more effective treatment options.

Research shows that the development of artificial intelligence technology brings with it many new opportunities and challenges. In recent years, generative models such as Generative Adversarial Networks (GANs) have become increasingly important for generating realistic images, sounds and text. In this review article, authors propose new GAN architecture enhanced with CLAHE algorithm to improve process of generating grayscale images. We employ CLAHE to preprocess the X-ray images, which enhances the local contrast and highlights the important features necessary for accurate diagnosis. The preprocessed images are then fed into a GAN architecture designed for image analysis classification. We evaluate our approach on a medical dataset of chest X-ray images and achieve a remarkable accuracy of 93.61%, significantly outperforming proposed method. The results demonstrate that the integration of CLAHE with GANs can effectively enhance the diagnostic performance of deep learning models in grayscale imaging.

In case of solving many classification problems, various machine learning models are used such as: KNN classifier, Decision Tree, Random Forest or even SVM but not only. In classification problems neural networks also have found broad applications, these networks help recognize and analyze photos. Additionally, they are utilized likewise to anomaly detection, such as detecting objects that differ significantly from the rest of the data in the data set. Generative Adversarial Networks deal with processes like: improving photo resolution, changing their colors, or creating typically new images. In this paper we will look at three different configurations of DCGAN models, especially at various types of generators and we will analyze the facial photos that were generated by each creator. We will show that one of the models that produced the most realistic images included downsampling and upsampling layers in the generator. Its architecture resembles that of a simple autoencoder. In addition, we will compare how every DCGAN architecture coped with producing a selected photo that was only in our collection of real photos.

As a result of the information analysis of various approaches to the types of interaction of robots in a group, the following tasks are solved: an environment model and the algorithm for mobile robot autopilot in the conditions of information uncertainty and a multi-agent simulation model and algorithm for focus group behavior of a team of robots in the framework of inter-agent communication were implemented. Based on the proposed simulation model the software has been developed that allows studying the effectiveness of group robots behavior algorithms.

This research explores the security challenges in leveraging Blockchain technology within the FinTech sector, particularly focusing on the control of digital trustworthiness in combating money laundering. Despite Blockchain’s promise in enhancing the security and integrity of digital transactions and identities, the study identifies critical security gaps in technical and operational security controls over FinTech domains. The findings emphasize the necessity for a concerted effort among researchers, technology developers, regulatory authorities, and financial institutions to address these challenges.

This study compares and evaluates different methods for malware detection. It specifically examines the effectiveness of ensemble learning, which combines predictions from multiple models (such as Decision Trees (DT), Support Vector Machines (SVM), and Random Forest (RF)) using a voting approach, against a advanced deep learning algorithm, the Deep Neural Network (DNN). The evaluation encompasses a rigorous assessment of these models using the widely recognized CIC-MalMem-2022 dataset. To optimize the model performance, a strategic feature selection process is employed, reducing the number of input variables and consequently streamlining training time. This consideration is pivotal in achieving efficient and practical detection solutions. The study covers both binary and multi-class classification scenarios, adding a layer of complexity to the evaluation process. By addressing multiple classification types, the research accounts for diverse real-world applications and extends the understanding of model behavior across different scenarios. In the context of performance evaluation, various metrics are employed to gauge the models’ effectiveness, including accuracy, precision, recall, F1-Score, and ROC-AUC. These metrics provide a comprehensive overview of the models’ predictive capabilities. Notably, in two distinct scenarios, the ensemble learning approach demonstrated substantial promise by attaining commendable accuracy rates, outperforming the DNN model.

In view of the fact that modern 21st century technologies have become greatly dependent on radio signals, it stands to reason that their security has moved to the front bon of priority. The paper under review is largely concerned with the popular types of radio hacking, it gives a detailed description and demonstration of some of the types and it also offers an in-depth analysis of GPS spoofing, including its risks and the measures that can be taken to prevent it. Another, main topic is ADSB system of ADS-B. This system periodically sends vital flight information units such as an aircraft location and an altitude. The work reviews both forms of spoofing technique in detail by noting their potential consequences for the navigational application safety and airspace integrity. In addition, it provides numerous case studies to demonstrate the changing landscape in navigation and aviation cybersecurity threats.

With the development of chess engines, cheating online has never been easier, resulting in a need for more robust and accurate detection systems. This paper presents a novel approach to chess cheater detection that combines conventional chess engines and neural networks to help identify which games are authentically played by humans and which show signs of extraneous intervention. By utilizing Stockfish to measure centipawn loss and its mathematical derivatives, we can measure deviations from typical computer-generated moves much like in conventional anti-cheat systems. Additionally, the neural network Maia, designed specifically to mimic human play, transmutes centipawn loss data to highlight deviations from human style. This dual-measurement system addresses the limitations of the given traditional anti-cheat systems, which face the issue of distinguishing between strong human players and those using engines. The collected metadata is analyzed using a sequential neural network, which identifies patterns of fair play violation. Our approach offers a robust solution for maintaining the integrity of online chess by accurately detecting and preventing cheating.

The work is about the development of a web-based website selection decision support system, which involves the development of the system’s data model, algorithm, and user interface. The work proposes evaluation criteria for websites based on search engine optimization metrics. The system’s algorithm ranks websites based on many criteria. This allows the user of the system to use the evaluation criteria offered by us and, in case of desire, to increase the number of evaluation criteria. The algorithm is partially expert because the evaluation criteria proposed by us are not evaluated by a human-expert, their value will be determined by the system for the corresponding website, and as for the evaluation criteria added by the user, it is evaluated by a human-expert. This system will enable users to select the best websites for their business quickly and easily, without excessive costs.

In recent decades, numerous algorithms have been proposed for extracting tables from PDF documents, with most designed as general-purpose table extractors. This paper introduces a novel algorithm specifically tailored for extracting order tables, addressing specific issues not covered by existing general-purpose extractors. Our approach, named OTE, identifies leading rows in an order table through a clustering algorithm and employs heuristics to recognize additional row lines and annotate columns. Through an evaluation of 115 order documents from customers of a medium-sized company, we demonstrate that: i) OTE surpasses general-purpose extractors, ii) accurately identifies over 95% of order tables in PDF documents, and iii) correctly identifies 81% of all listed article IDs, even when included in the article description.

Process mining, introduced by Aalst in 2004, is a data analytics technology designed to extract process-related information from historical event logs. This field has seen significant growth, offering numerous tools and applications that provide data-driven solutions and support for process improvements. Modern business processes, including management, modeling, and analysis, are primarily conducted through process-aware information systems such as Customer Relationship Systems, Accounting and Finance Management Systems, and Enterprise Resource Systems, which collect process events in various event logs. Visualization of process mining can be very valuable to the decision-making process. Visual representations of mined process data allow participants of all sides to easily interpret complex information, identify patterns, and detect anomalies. By transforming abstract data into intuitive visual formats, decision-making team can gain genuinely worth insights faster and more accurate. This facilitates better understanding of process performance, helps detect inefficiencies, and supports the optimization of business activities. Effective visualizations also enhance communication among team members and other part participants, ensuring that data-driven decisions are well-informed and aligned with organizational goals. Thus, process mining visualization is a crucial tool for leveraging data to drive strategic and operational improvements. The main purpose of the paper is to present several process mining visualization examples resulting from the analysis of historical event logs.

This article discusses the importance of effective risk management in Agile projects. The Agile project management philosophy emphasizes open communication, flexibility in product development, and rapid market delivery. Unlike traditional waterfall methods, Agile iterative approach reduces risks by allowing on-going progress assessment. However, Agile alone may not effectively manage risks, especially those stemming from human factors and environmental influences. Existing project management tools lack comprehensive risk identification and management solutions, particularly concerning diverse project components and stakeholder engagement. Decentralization in Agile projects can lead to architectural risks due to documentation inconsistencies and client disengagement. To address these challenges, this article proposes a methodology for identifying risks of Agile project associated with artifacts and attributes of it. By regularly updating attributes’ data, potential risks can be unveiled and addressed early, enhancing project team effectiveness and stakeholder clarity through transparent reporting.

Dating manuscripts, particularly when dealing with handwritten materials, poses a distinct challenge, especially compared to identifying authors in anonymous holographs. The wide array of differences in handwriting styles among various authors, coupled with the tendency for these differences to evolve over time, can often obscure more subtle nuances within the handwriting of a single author. In response to this challenge, our research endeavors to explore a diverse set of methodologies and approaches aimed at precisely dating the undated holographs of Galaktion Tabidze, a prominent Georgian poet from the 20th century. This article outlines two distinct approaches utilized in our study, delving into the experiments conducted to assess the efficacy of each proposed dating method. Through the investigation of these methodologies, our objective is to offer valuable insights and enhance the accuracy of dating historical holographs, thus contributing to a deeper understanding of Tabidze's work and the broader historical context in which it was created.

This paper investigates the latency challenges in field-edge systems within transport monitoring environments, with a focus on collision detection and object tracking. Leveraging a field-edge network architecture, we explore the integration of image preprocessing at the field device level and complex object detection and tracking models at the edge device level. Our approach is focused on optimizing the balance between processing speed and accuracy, addressing the critical need for real-time responsiveness in dynamic scenarios such as smart city infrastructure and autonomous navigation. We build on the foundation of existing object detection algorithms, i.e. YOLO, and introduce a batch processing strategy with parallel threads to mitigate latency issues inherent in sequential processing frameworks. Our system is tested using a YOLOv8 model trained on a custom Traffic Object Detection dataset, demonstrating the ability to maintain high processing speeds, with performance exceeding 30 frames per second, without sacrificing accuracy. The approach combines advanced object detection and tracking algorithms with network optimization. Our system tests show positive results in reducing latency. This work sets a benchmark for the deployment of field-edge systems in transport monitoring and contributes to future advancements in smart transportation and autonomous systems.

Text simplification involves reducing complexity while retaining important information. This is important for improving accessibility for a wide range of readers, including those with cognitive disorders, non-native speakers, and children. We report experiments on text simplification for Lithuanian with a focus on simplifying texts, written in an administrative style, which is not easy to understand for the general public. We chose mT5 as a foundational model and fine-tuned it in 3 different ways. We evaluated models’ outputs with ’unseen’ data and automatic evaluation metrics accompanied by qualitative analysis. Though automatic evaluation scores were higher for the model fine-tuned with translated data, qualitative analysis revealed that the model that was fine-tuned with specifically for this task prepared Lithuanian data performed simplification task significantly better.

Proportional representation is one of main families of voting methods for electing several candidates. It is further subdivided into party list based proportional representation and single transferable vote (STV). It is known that it is possible to simplify the calculations needed for implementation of single transferable vote by using one of the methods based on party lists - largest remainder method, if the preference lists are made of relatively long non-intersecting party lists (possibly listing one party list after another in the preference list, but keeping the order in each party list). This paper discusses a way to produce the simulated party lists that would make it possible to use the same algorithm without explicitly constructing preference lists from the party lists.

Hybrid learning, which integrates in-person and online learning experiences, presents specific delivery challenges that affect the quality and effectiveness of instruction. These challenges vary depending on the context and available technologies. This paper explores different issues in delivering hybrid learning, focusing on synchronous and asynchronous learning methods, technology integration, and finding the right balance between in-person and online interactions. It also highlights technical issues such as internet connectivity, platform compatibility and audio and video quality. In addition, it discusses the advantages of hybrid learning, including cost-effectiveness and increased accessibility, as well as challenges such as reduced face-to-face interaction and technological differences between students. Addressing these issues requires comprehensive support, equal access to resources, and effective training for educators and students. Ultimately, the effectiveness of hybrid learning depends on a well-designed curriculum, teacher training, and seamless integration of technology.

The online learning activities have gained prominence during the past three years, and an appreciable amount of significant learning components were adapted for online learning patterns. Pandemic enhanced the conceptual and practical preoccupations concerning the online learning processes, with an emphasis on the overall development of the networking techniques, software tools and services. There are various learning materials, such as open papers, books, coding examples, dictionaries and online trainings, which are dedicated to specific didactic processes, or transmitted through animation-based approaches. Thus, the existing scientific literature proposes new machine learning and natural language processing-based mechanisms, which are related to, for example, materials dedicated to learning foreign languages. The consideration of such learning methods may be efficiently used in order to enhance the study of structured learning scopes, such as computer science, mathematics, or other subjects. Consequently, the relevant learning processes are re-structured and enhanced. This work addresses approaches that relate to the combination of classical learning techniques like recognition, intercalation, recall time, and written notes, with proper tools and improved educational processes. This paper presents an integrated online learning paradigm, which is validated considering the didactic interaction with relevant student groups. Additionally, this article discusses the real-world benefits of the proposed learning model, and on the open research questions that will be approached by future research projects.

Innovative educational technologies, the integration of Massive Open Online Courses methodology (MOOC), Challenge-Based Learning (CBL), and Virtual Assistant methodologies in Big Data course represent a dynamic evolution in pedagogical approaches. MOOCs offer scalable access to high-quality educational content, enabling learners to engage with Big Data concepts flexibly. CBL fosters critical thinking and problem-solving skills by immersing students in real-world scenarios relevant to Big Data analysis. Virtual Assistant methodologies leverage artificial intelligence to provide personalized learning experiences, enhancing student support and interactivity. This integrated approach not only cultivates a comprehensive understanding of Big Data but also prepares learners for the demands of a data-driven world. The authors are discussing the methodology and the effectiveness of the implemented course.

AI technologies offer transformative potential for tutoring and achieving learning outcomes, addressing these challenges is critical. Collaboration among educators, technologists, policymakers, and stakeholders is essential to create effective AI-driven educational solutions. A paper presents the case on artificial intelligence technologies using for tutoring and achieving learning outcomes. We analize existing best solutions for educational practice in tutoring which is still challengeable for many academics and described how technologies can improve or support learning process or achievement of the learning outcomes. Moreover, we identified a crucial tutoring points to address technical, ethical, and pedagogical considerations to maximize their effectiveness and ensure positive learning experiences for students by using chatbots in education for tutoring.