This study presents a log-based learning analytics pipeline for quantifying user engagement in Moodle, demonstrating how event log data can be transformed into analyzable interaction patterns through extraction, anonymization, categorization, and statistical modeling. The approach applies distributional testing using chi-square statistics and Cramér’s V to identify structural differences in user activity. As a case study, the method was implemented in an English for Specific Purposes (ESP) course, comparing a control group following a traditional LMS configuration (N = 40) and an experimental group using a gamified Moodle environment (N = 40). Results indicated that the gamified configuration generated significantly higher frequencies of system-level and assessment-related interactions, as well as more sustained activity across the instructional period, while the control group relied primarily on static content access and exhibited declining participation over time. Engagement was operationalized through event frequencies, capturing observable behavioral differences rather than cognitive or learning outcomes. Beyond the educational setting, the study illustrates how reproducible event-log analytics can be used to detect behavioral shifts in technology-supported environments, offering a methodological template that is potentially transferable to similar contexts in applied computer science.
Data normalisation is a critical preprocessing step for machine learning, especially for high-dimensional, small-sample datasets such as those encountered in microarray analysis. This study comprehensively investigates the impact of eight distinct normalisation methods, including Vector Normalisation (L2 Normalisation), Quantile Normalisation (Gaussian and Uniform), Maximum Absolute Scaling, Z-score, Min-Max, Power Transformation, and Robust Scaling, on the classification performance of microarray data. Using an Extreme Learning Machine (ELM) as the classifier, the research evaluates performance across three leukaemia datasets with varying numbers of classes, namely 2, 3 and 4 classes. The results demonstrate that Vector Normalisation consistently outperforms all other methods. In the 2-class scenario, it achieved the highest accuracy (87.50%) and F1-score (87.08%). Although unnormalised data showed a similar average accuracy, Vector Normalisation proved empirically superior due to its significantly lower standard deviation, which is10.89, indicating a more stable and reproducible model. This stability became even more critical in the 3-class scenario, where overall performance declined, but Vector Normalisation still led with 61.67% accuracy and 52.18% F1-score, while other methods, particularly simple scaling techniques like Min-Max, showed a sharp drop and extreme instability. In the 4-class scenario, a performance rebound occurred, and Vector Normalisation maintained its top position, achieving 72.92% accuracy and an F1-score of 66.17%. The findings confirm that Vector Normalisation is the most effective normalisation method for microarray data, delivering both high performance and superior stability across varying levels of class complexity.
Additive manufacturing is widely used for prototyping and producing functional parts. With the growing capabilities of industrial robots, robotic additive manufacturing is becoming an attractive alternative to conventional 3D printing. Robots enable the fabrication of large-scale, structurally complex, and non-planar components that exceed the limitations of traditional printing. However, the flexibility of robotic systems comes with increased complexity in process control – particularly in the selection of appropriate printing parameters, which is critical for ensuring the quality and stability of the printed parts. This paper addresses the need for a systematic approach to parameter selection in robotic 3D printing to ensure optimal process performance and part quality. First, control software to manage and execute the printing process was developed. Secondly, the impact of changes in the industrial robot TCP's velocity and orientation on the quality of manufactured parts was investigated. Furthermore, to optimize the selection of process parameters, the TOPSIS multi-criteria decision-making method was employed. The presented approach provides a methodology for parameter selection and optimization in robotic 3D printing.
The paper presents a new approach to optimising cluster structure by selecting servers to meet specified performance requirements while minimising costs. Modern applications are placing increasing demands on performance and are critical elements of business operations. As a result, the operation of such applications increasingly relies on server clusters. Selecting the type and number of servers is not a trivial task. The problem is further complicated by the widespread use of layered application architectures, which means that different hardware solutions may be optimal for handling different layers. The article proposes a technique that uses the Tabu Search heuristic in conjunction with a BCMP-based application model. An optimisation algorithm for the cluster structure is presented in two versions: minimising the solution cost while meeting performance requirements, and maximising performance while meeting budget constraints.
Water Vortex Hydro Turbines are a promising renewable energy solution for ultra-low-head applications, with performance strongly influenced by blade geometry in converting vortex-induced momentum into mechanical energy. This study numerically investigates the hydrodynamic characteristics including velocity distribution, air-core formation, and pressure gradient to determine the optimal blade angle using transient Computational Fluid Dynamics (CFD) with the Realizable k-ε turbulence model. A six-bladed turbine was analysed at blade angles of 75°, 90°, and 105° under a constant inlet velocity of 1 m/s. The results demonstrate that the 90° configuration delivers the best overall performance, with the maximum velocity increasing from 3.096 m/s (75°) to 4.376 m/s (90°), representing an improvement of approximately 41.4% and indicating a significantly stronger and more stable vortex structure. Although the 105° configuration reaches the highest peak velocity of 4.657 m/s, severe flow separation reduces flow stability and limits effective energy transfer. In terms of pressure, the 90° configuration produces the most favourable and symmetric gradient, with a maximum of 5777.397 Pa and a minimum of 3.154 Pa, compared to the more distorted distributions observed in the 75° and 105° cases (5741.087 Pa and 5892.066 Pa, respectively). Additionally, the 90° configuration generates a more compact and stable air-core, reducing hydraulic losses and enhancing fluid–blade interaction. Overall, the 90° blade angle provides the optimal balance between velocity magnitude, vortex stability, and pressure distribution, making it the most effective configuration for maximizing energy conversion in ultra-low-head water vortex hydro turbines.
The prescription and documentation of controlled medications require strict regulatory compliance and high transcription accuracy to prevent medication errors and ensure traceability. In many hospitals, these processes are still performed manually, increasing the risk of transcription errors, administrative delays, and non-compliance with regulatory standards, particularly for medications classified under fractions II and III of the Mexican General Health Law. Addressing this challenge requires intelligent systems capable of accurately transcribing and structuring medical prescriptions from spoken language. This study presents the design and development of an Automatic Speech Recognition (ASR) system integrated with Natural Language Processing (NLP) to support the generation and transcription of controlled medication prescriptions. The system architecture was developed following an analysis of the clinical workflow for medication requests, management, prescription, and transcription, conducted in collaboration with healthcare professionals from the hospital's Pharmacovigilance Department in Puebla, Mexico, and aligned with hospital operational standards. The methodology involved evaluating and fine-tuning three ASR models to improve transcription accuracy for medication names, dosages, and prescription instructions. NLP techniques were subsequently applied to identify and structure key prescription entities, ensuring compliance with national health regulations. Among the evaluated models, the Wav2Vec2 architecture developed by Jonatas Grosman demonstrated the best performance and was selected for implementation. Experimental results show that the optimized ASR model achieved a Word Error Rate (WER) of 6.30%, a precision of 94.72%, a recall of 91.73%, and an F1-score of 93.22%. These results demonstrate the effectiveness of the proposed approach in improving transcription accuracy while reducing false positives in prescription generation. The proposed system highlights the potential of ASR–NLP integration to enhance efficiency, accuracy, and regulatory compliance in hospital pharmacovigilance processes.
The role of business operations driven by information technology systems is crucial, particularly the use of Enterprise Resource Planning (ERP) systems. Within organizations, both operational staff and management engage with ERP systems, which introduces potential vulnerabilities to operational errors or fraudulent activities. Consequently, auditing application controls becomes a matter. This study conducted a Systematic Literature Review (SLR) to investigate the scope of internal audit, with a focus on application control and associated risks. Based on the SLR results, an application controls audit framework for ERP is proposed. It consists of ten essential controls that must be implemented, including the software development process, access control, input control, process control, output control, change control, incident control, legal and ethical control, information security risk management, and continuity control. The framework evaluation, based on two case studies, demonstrated high effectiveness and received positive feedback from IT auditors, auditees, management, and executive boards.
Under the long-term action of train loads and complex environmental conditions, the surfaces of railway tracks are prone to defects such as cracks, spalling, and pitting, which seriously threaten the safety of railway operations. Semantic segmentation can achieve pixel-level positioning and morphological characterization of defects. However, existing methods still struggle to model strongly directional structures and multi-scale defects while maintaining a balance between accuracy and efficiency in rail-surface inspection. To address the above issues, this paper proposes a lightweight semantic segmentation network for railway track surface defects (SFAB-Net) based on spatial fusion and adaptive bottleneck feature enhancement. This network effectively characterizes the features of slender cracks along the rail direction using the direction-sensitive Spatial-Fusion module and combines them with the simplified spatial pyramid pooling module to achieve multi-scale context aggregation. In the decoding stage, an adaptive feature reconstruction mechanism and spatial-channel joint attention are introduced to enhance multi-scale feature fusion and suppress background interference. Experimental results on the NEU-DET dataset and a self-built rail surface image dataset show that SFAB-Net outperforms several representative methods in segmentation accuracy and robustness, and has strong potential for engineering applications.
Organizations in transitional economies face significant challenges in implementing effective knowledge management systems owing to infrastructure limitations, post-conflict instability, and cultural barriers that prioritize tacit knowledge over formal documentation. This study addresses this critical gap by developing and evaluating a context-sensitive Cloud-based Knowledge Management System (CKMS) specifically designed for Iraqi firms operating in resource-constrained conditions. Employing a Design Science Research (DSR) methodology integrated with an explanatory sequential mixed-methods approach, we collected data from 350 knowledge workers across ten organizations and conducted structural equation modeling to examine the relationships between CKMS implementation, knowledge management capabilities, and organizational innovation outcomes. The findings demonstrate that CKMS significantly enhances all four knowledge management capabilities—acquisition, storage, sharing, and application—with particularly strong effects on knowledge storage (β = 0.72, p < 0.001) and knowledge application (β = 0.70, p < 0.001). These enhanced capabilities subsequently drive substantial improvements in organizational innovation, with process innovation showing the strongest impact (β = 0.56, p < 0.001), followed by product innovation (β = 0.42, p < 0.001) and business model innovation (β = 0.38, p < 0.001). Leadership support and organizational culture emerge as critical moderating factors that amplify the effectiveness of CKMS. This study extends Dynamic Capabilities and Resource-Based View theories by demonstrating how context-adapted technological solutions can function as strategic resources in transitional economies, providing a validated framework for digital transformation that prioritizes accessibility and cultural fit over technological sophistication, thereby offering actionable insights for organizations and policymakers operating in similar post-conflict settings.
This study presents an automatic detection system for suspicious facial objects in neutral automated teller machines (ATMs), using and comparing the deep learning architectures YOLOv8 and Faster R-CNN. A dataset was built from real ATM surveillance videos and complementary images, from which frames were extracted and annotated with masks, hats, and glasses. Both models were trained under the same preprocessing pipeline and evaluated using standard object detection metrics such as precision, recall, F1-score, Intersection over Union (IoU), and mean Average Precision (mAP), in order to analyze their performance in real surveillance conditions. The results show that YOLOv8 achieves higher precision, reducing the generation of false positives, while Faster R-CNN demonstrates higher recall and superior mAP@0.5 values in several classes, indicating greater sensitivity to partially visible objects. A decision-making logic was also integrated to automatically classify each scene as NORMAL or SUSPECT, based on the combined presence of facial-occluding elements. The implementation demonstrates that computer vision systems can complement security mechanisms in neutral ATMs by providing early detection of potential risks and enabling real-time remote monitoring.
Plagiarism is a common issue in programming education, and the issue exacerbates with the emergence of Generative Artificial Intelligence (GAI). Plagiarism acts can be disguised with GAI, resulting in pervasive, consistent changes across the entire program. We present a programming plagiarism detector dedicated to GAI disguises. It not only relies on program similarities but also on GAI characteristics. GAI has its own way of writing programs. Our plagiarism detector employs 23 features. Five of them are related to structure (program similarities) while the rest are associated with GAI characteristics (the use of list comprehension, recursion, etc). It features seven machine learning models to choose from: Logistic Regression, Random Forest, XGBoost, LightGBM, CatBoost, Voting Classifier, and Stacking Classifier. According to our evaluation of 6344 instances from the machine intelligence course, Stacking Classifier achieves the highest performance, with 89.17% accuracy, 88.94% precision, 89.17% recall, and 88.77% F-score. It outperforms similarity-based plagiarism detectors (which serve as the baseline) by a factor of 2 in most metrics. All structural features (program similarities) are considered important by our machine learning models, accompanied by several GAI-characteristic features. The prominent GAI characteristics are the use of list comprehension, recursion, and branching condition statements without parentheses.
Swarm robotics is a particularly promising approach for autonomous exploration in complex and uncertain environments, with applications ranging from environmental monitoring to hazardous-area inspection. A major challenge lies in optimising robot trajectories to minimise travel distance while ensuring comprehensive and effective coverage of the exploration area. In this context, we propose a hybrid path-planning framework that combines the SARSA Reinforcement Learning algorithm with the ACO approach, drawing inspiration from collective coordination mechanisms in nature, particularly the use of pheromones as a medium for self-organisation. This framework leverages both individual learning and swarm intelligence in a complementary manner, thereby enabling more robust, scalable, and efficient exploration. A comparative analysis of the two methods was conducted to identify the most effective approach for optimising robot trajectories while minimising energy consumption. In this process, robots take into account obstacle avoidance, whether obstacles are traversable, using either pheromone-based environmental marking or reinforcement learning strategies. Simulation results demonstrate the effectiveness of a hybrid model that integrates SARSA with ACO, significantly enhancing trajectory quality and exploration coverage. However, they also reveal that increasing the environment size substantially increases the total travel distance and slows SARSA convergence due to the expansion of the state space. To overcome this limitation, future work will explore neural network–based value function approximation, which is expected to improve generalisation and accelerate convergence in large-scale scenarios.
Content-based image retrieval (CBIR) systems play an important role in many applications, including object recognition, digital forensics, and biomedical research. Handling large volumes of digital images in our daily lives requires the development of an efficient CBIR system. In this paper, a comprehensive and unified approach for image retrieval has been introduced. It brings together deep learning, feature reduction, and clustering optimization in a way that hasn't been explored before. The pre-trained VGG16 model has been used to extract rich, high-level features from images. To make the process more efficient and focused, Principal Component Analysis (PCA) is applied to reduce the number of features while retaining the most important information with minimal runtime. What sets our work apart is the use of hybrid K-means clustering with Particle Swarm Optimization (PSO). Instead of relying on random initialization, PSO helps find better starting points and improves the overall clustering by guiding K-means away from local minima. These well-known techniques are combined into a single, coherent framework called VGG16-PCA-PSO-K-means. The effectiveness of the proposed method has been evaluated on the Corel 1K and UC Merced Land Use datasets using mean average precision (mAP), clustering purity, recall, F-score, NDCG, and runtime. The results of the experiments conducted indicate that the proposed system achieves higher precision and clustering purity than state-of-the-art methods. The improvement ranged from 5% to 18% across different feature set sizes, with mAP@10 of 97.5% for the 10-class retrieval of the Corel 1K and mAP@10 of 96.7% for the 21-class retrieval of the UC Merced Land Use datasets, using only 30 features. Furthermore, the results of the Kendall's W and Friedman tests confirmed that the (VGG16-PCA-PSO-K-means) model achieved a higher rank than other methods in the literature, with a significant difference, highlighting its effectiveness and robustness.
The competitive process for price-setting in digital markets is being dramatically altered through the use of autonomous artificial intelligence (AI) to make pricing decisions on behalf of humans. These systems operate autonomously and interact with each other in continuous cycles. They react in real time to market data and adapt their pricing strategies accordingly. This research analyzes the effects of various combinations of market transparency and algorithmic autonomy on price behavior, the competitive process, and consumer welfare outcomes. The analysis is conducted using a controlled simulation model that compares four pricing regimes: human-supervised, fully autonomous, isolated, and mediated. The results show that the degree of market transparency and the degree of platform oversight of AI decision-making have a far greater impact on the market's final outcome than the level of algorithmic autonomy. Some configurations increase efficiency and profit while increasing the risk of coordination failure, volatility, and concentration. In addition, the findings demonstrate inherent structural trade-offs between market efficiency, market stability, and competitive processes in markets where AI is used as an autonomous decision-making agent. These results also highlight the limitations of attempting to regulate AI through intent-based regulations and provide insights into how autonomous AI decision-making agents alter the structure of digital markets.
Unmanned Aerial Vehicle (UAV) imagery, augmented by advanced deep learning architectures, has become an integral approach to the automated inspection and structural monitoring of high-voltage power grids. This study evaluates the practical applicability and speed-accuracy trade-offs of single-stage versus two-stage object detection models for identifying critical power grid components. Specifically, two highly optimized networks were developed and empirically validated: HVE-YOLO11, which uses the latest YOLO11 architecture enhanced with spatial attention mechanisms, and HVE-MASK-R-CNN, which uses a rigorous ResNet-101 Feature Pyramid Network (FPN) backbone. Leveraging a newly curated, diverse dataset of 51,800 augmented images capturing six distinct equipment classes under fluctuating meteorological conditions, the models were evaluated for Mean Average Precision (mAP) and computational throughput measured in Frames Per Second (FPS). Empirical results demonstrate that the single-stage HVE-YOLO11 shatters the traditional speed-accuracy dichotomy, significantly outperforming the two-stage model in both inference velocity (162.9 FPS versus 75.8 FPS in intensive benchmarking) and spatial accuracy (an mAP@0.5 of 0.972 compared to 0.855). These findings provide actionable, highly quantified benchmarks for deploying real-time, AI-driven diagnostic systems on hardware-constrained edge-computing UAV platforms.
This paper investigates the applicability of offline-first strategies for workload placement in multi-cloud environments, focusing on integrating on-premise resources. The authors analyze the evolution from traditional on-premise IT infrastructures to hybrid and multi-cloud models, highlighting the growing complexity of interoperability, cost management, and vendor lock-in. Sky Computing is presented as a paradigm that abstracts cloud resources across providers via an intercloud broker, enabling dynamic, vendor-agnostic resource allocation and service migration. The paper explores the challenges of secure connectivity, data gravity, and latency in distributed environments. It discusses the role of Software Defined Networking (SDN) and Secure Access Service Edge (SASE) in facilitating seamless and secure integration of on-premise and cloud workloads. A key contribution is introducing the SKY CONTROL framework, designed to address the specific needs of small and medium-sized enterprises (SMEs) by providing cost and risk management, infrastructure transparency, and support for geographic workload distribution. The study concludes that Sky Computing and offline-first strategies offer significant potential for enhancing flexibility and resilience in multi-cloud environments. However, further research is needed to develop standardized methodologies for workload placement and address open interoperability and security challenges.
Developing software demands strict compliance with high quality and reliability criteria. When dealing with safety-critical systems, where failures can cause severe consequences, these requirements become even more critical, making the testing process especially important and costly in terms of resources. The automated generation of test cases has arisen as a potent method for identifying a greater number of defects while diminishing the need for manual intervention. In particular, Dynamic Symbolic Execution (DSE) has been presented in the literature as an effective technique for the automatic generation of test cases in programs written in C/C++. This work presents a study on its practical applicability in open-source C/C++ projects, using the tools KLEE and gcov to evaluate its effectiveness and the challenges encountered. Additionally, the advantages and limitations of this technique in real-world settings are discussed, along with potential strategies to enhance its applicability.
Climate-related risks have recently emerged as critical factors in financial risk management, resulting in significant demand for reliable and standardized sustainability data. In response, the European Union (EU) has established regulatory frameworks mandating corporations of varying sizes to disclose key sustainability indicators. These developments facilitate improved quantitative risk management through consistent and accessible sustainability data published by corporations in their respective reports. However, reporting from diverse and heterogeneous data sources necessitates structured methods for storing and organizing domain knowledge, enabling efficient data access and retrieval. To address these challenges, we introduce an ontology designed to function as a schema for constructing a Knowledge Graph (KG) that integrates data essential for assessing climate risks within financial institutions’ loan portfolios. Following established ontology modeling practices, our approach reuses and extends existing ontologies, ensuring alignment with the latest EU reporting standards specified in the Corporate Sustainability Reporting Directive (CSRD). This ontology, which is made publicly available, supports the assessment of transition climate risks, e.g., through a portfolio temperature alignment framework and further enhances capabilities for information extraction, identification of data gaps, and analysis of data integrity within sustainability reports.
Forecasting time series is a classical challenge in time series analysis. Applying machine learning based forecasting on the biomedical electrocardiogram (ECG) signals is not common but vital in predicting the overall health assessment of an individual. In this study, we forecast ECG signals using temporal fusion transformers (TFT) for the very first time. We use the renowned publicly available dataset, PTB-XL, to forecast all of the 12 leads. Different hyper parameters are used with temporal fusion transformer obtain optimal results. We downsample and normalize our dataset in order to reduce computational complexity as part of pre processing. Then it is fed into a TFT with different hyperparameter settings. We get an average root mean square error of 0.061 and mean absolute error of 0.128. The promising results encourage the application of TFTs for forecasting of ECG signals specifically for a better explainability emphasizing the research in explainable artificial intelligence (XAI).
Legacy code is an inherent part of industrial software projects, often resulting in complex, tightly coupled systems that challenge the integration of modern testing techniques. Unlike in academic contexts, where tools are developed with well-scoped objectives and simplified scenarios, industrial environments demand higher levels of robustness, performance, and adaptability. Refactoring, therefore, becomes a crucial process when adapting existing testing tools to support legacy code in production settings. This paper presents the refactoring of a testing tool originally developed for academic use, with the aim of making it compatible with real-world industrial C++ codebases. The project, conducted in collaboration with a naval industry partner, involved reengineering the tool’s architecture to improve readability, modularity, and execution performance. Through a case study approach, we illustrate the advantages achieved by systematically applying refactoring principles, resulting in a more robust and adaptable tool capable of supporting automated software testing in complex environments. Additionally, we provide a practical refactoring guide based on our experience, offering valuable insights for teams facing similar adaptation challenges in the future.
