Imbalanced class distributions in Internet of Things (IoT) attack datasets limit the ability of machine learning (ML) models to detect minority intrusions and lead to high false-negative rates. This study investigates the effectiveness of the Synthetic Minority Oversampling Technique (SMOTE) in improving multi-class attack detection across seven IoT devices from the TON_IoT dataset. Four ML algorithms —K-Nearest Neighbor, Support Vector Machine, Logistic Regression, and Multi-Layer Perceptron— are evaluated both with and without SMOTE using accuracy, precision, recall, and F1-score.Results show that SMOTE substantially improves minority-class detection, particularly in devices with severe imbalance. On the Garage Door dataset, KNN accuracy increases from 0.952 to 0.982, and MLP from 0.555 to 0.978, while SVM exhibites modest gains. The largest recall improvement is observed for Modbus with KNN, rising from 0.52 to 0.84.Overall, SMOTE enhances IoT intrusion detection without degrading majority-class performance, although its impact varies across algorithms and device characteristics. Future research is planned to explore hybrid resampling and deep learning-based approaches to further improved detection under extreme imbalance.
Heart failure (HF) is a major global health issue and a leading cause of hospitalization as well as mortality, specifically among the elderly. Accurate prediction and continuous monitoring of HF patient survival are essential for effective clinical decision-making. Therefore, this study aimed to develop a web-based software system that incorporates multi-algorithm, Naive Bayes (NB), Random Forest (RF), K-Nearest Neighbor (KNN), Support Vector Machine (SVM), and Neural Network (NN), to predict patient survival using 12 clinical variables. The software enabled efficient diagnosis and monitoring by generating predictive outcomes based on patient data. In the evaluation phase, the NB algorithm achieved the highest accuracy of 88.21%, followed by RF (85.11%), NN (82.36%), SVM (76.11%), and KNN (62%). The major originality of this study resided in combining multi-algorithm into a unified diagnostic platform, offering practical and low-cost decision support for healthcare providers. In contrast to previous studies that focused on developing standalone model, this analysis produced a functional application directly implemented in clinical environments. Relating to the analysis, future studies should aim to expand the dataset, incorporate real-time hospital data, and use explainable AI to improve interpretability, scalability, as well as trustworthiness in medical decision support systems.
To address the growing need for wireless communications energy efficiency, this paper proposes a new multi-agent reinforcement learning (MARL) approach to cooperative data offloading in heterogeneous cellular networks. This research is among the first to employ MARL to this extent, and it offers an end-to-end solution that combines cellular, Wi-Fi, and device-to-device (D2D) communications and considers practical network environments like user mobility and channel conditions. We formulate the offloading problem as a Markov Decision Process (MDP) with correct models of energy consumption and network conditions. The deep Q-network (DQN)-based MARL algorithm allows user equipment (UEs) to learn collaborative strategies for optimizing overall energy consumption and timely offloading of data. Simulations compare MARL against greedy, random, and independent Q-learning baselines in low and high mobility regimes. Experiments show that MARL saves energy by as much as 40% over random offloading and 16.6% over greedy offloading, as well as enhancing average delay, throughput, and fairness. Convergence of the learning rate of the algorithm is within 1000 episodes, and sensitivity analyses confirm its performance across a range of user density and data size settings. Furthermore, the MARL framework accommodates dynamic network conditions and provides an adaptable solution for network operators to maximize performance and sustainability for existing and future wireless networks. The suggested MARL framework still performs better in terms of delay, throughput, and fairness, while at the same time catapulting energy savings over the greedy offloading approach by 9.4% to 12.8% under the very realistic 3GPP-compliant HARQ, adaptive modulation, standardized power control, and urban SLAW mobility models settings. With the extended state/action spaces and realistic 3GPP+SLAW conditions, MARL has an energy savings of 11.7%–14.2% over greedy offloading while still having the leading delay, throughput, and fairness.
The Saga pattern guarantees eventual consistency in microservices but depends on passive post-failure compensation, resulting in resource inefficiency and possible data discrepancies. This study introduces the ”Proactive Saga Failure Avoidance (PSFA)” system, which incorporates deep learning-based health evaluation into the Saga orchestrator. The platform employs an LSTM model to monitor real-time service metrics, enabling the prediction of imminent failures prior to transaction execution. Upon detection of high risk, the transaction is proactively terminated to prevent erroneous computation and intricate rollbacks. Experimental findings obtained in a controlled environment indicate that PSFA detects failure precursors with an accuracy of 1.00 and a recall of 0.99. This method provides an innovative alternative to improve the reliability and efficiency of distributed transactions in microservice settings.
The Internet of Things (IoT) has experienced rapid growth, taking advantage of advanced data processing. While complex event processing (CEP) engines have positioned themselves as a good solution for cloud and fog data processing, there are no CEP solutions designed to run on microcontrollers (MCUs) at the edge. This paper presents MicroCEP, a software architecture for MCUs that enables real-time filtering and correlation of sensor data, both locally obtained and received through message queues at the edge. MicroCEP provides (i) a lightweight CEP architecture for MCUs, (ii) a pattern language to specify complex event patterns, (iii) a REST API for dynamic threshold updates, and (iv) a methodology for designing and deploying MicroCEP-based systems. The solution facilitates the detection of situations of interest and the triggering of actions directly from the MCUs, enabling autonomous decision-making close to the sensors. A proof of concept implementation using RISC-V MCUs provides evidence that MicroCEP can deliver CEP capabilities at the edge on resource.constrained devices.
he objective of this study is to analyze the acceptance and ethical conditions required for the implementation of non-invasive monitoring systems for older adults. Using a mixed-methods design, it integrates focus groups with older adults and a questionnaire with future social and healthcare professionals. The results show a high level of acceptance of wearable technologies that do not use cameras or microphones, highlighting their discreet nature, protection of privacy, and contribution to autonomy and safety. Participants valued person-centered design, transparency in data use, and technologies that complement—rather than replace—human support. The main perceived barriers were cost, lack of training, and the digital divide. These findings indicate that wearable devices can support aging in place when developed through participatory and ethical approaches that respect dignity and promote trust. The study concludes that successful implementation requires user involvement, clear communication, and sustainable strategies for integration into social and healthcare services.
The development of sixth-generation (6G) mobile communications aims to deliver super-fast connections at terahertz frequencies, surpassing those of 5G. Nonetheless, these improvements come with challenges of data leakage and packet misinsertion, and they place greater pressure on secure and efficient routing. An Energy Optimized Network Route Cluster Bandwidth (EONRCB) is an Enhanced Service Data Transmission (ESDT) model proposed in this paper to circumvent these problems. The model contains several significant factors, such as Transmission Node Support Weight (TNSW), which evaluates network traffic and petite access control, which provides packet fragmentation and a secure route. Then, the Service Level Route Count Rollback Node Aggregator (SLR-CRNA), which validates active route node interaction. Further, the Sleeper Node Controller improves the consistency of node operations, whereas Slicing Neighbour Node Path Routing (SN2PR) introduces consistency in the source-to-destination connectivity within edge networks. Additional optimisation can be achieved through functionality such as Priority Cycle Tags (PCT) and Recursive Scheduling Time (RST), which provide mechanisms to improve overall RTPS throughput and node scheduling by minimising latency via a depth-first task alignment scheme. The suggested architecture achieves higher energy efficiency, reduced congestion, and a longer network lifespan. The performance of the proposed 6G framework will be validated against existing models to assess its effectiveness.
This study explores the application of predictive modeling to the TensorFlow project, a prominent open-source machine learning library. Our research aims to develop and validate predictive models that can accurately forecast issue closures, thereby enhancing project management decisions. We systematically collect data from the TensorFlow GitHub repository through data crawling using a self-built python script, focusing on key metrics such as commit history, issue tracking, and contributor metrics. The data comprises 120,000 commit messages and 25,053 issue tickets. We apply descriptive analytics to identify insights into the project's development dynamics and implement nineteen models using various machine learning techniques, including linear regression, ridge regression, and ensemble methods. The final model, based on Lasso regression, achieves high accuracy with the lowest Mean Squared Error (MSE), the Average Cross-Validation MSE and highest R-squared value compared to other models. The final model also utilizes advanced feature engineering techniques, such as lagged features and cyclical time transformations, which help enhance the model's performance. Applied time-aware validation shows the need to further refine the evaluation design to enhance temporal modeling.
Traditional resume screening is manual, slow, and susceptible to bias, and it struggles to keep pace with today’s application volumes. This paper presents a dual-engine, AI-powered resume screening system designed for transparency and reproducibility. The primary (classical) pipeline encodes resumes and job descriptions using Sentence-BERT (SBERT), computes a resume–job match score via cosine similarity, classifies candidates into 25 job categories using XGBoost, and provides model interpretability through SHAP. In parallel, a prompted large language model (LLM) baseline (GPT-4o/4o-mini) outputs a match score and predicted category for comparative analysis. A Streamlit-based interface integrates both engines to support recruiter workflows and human-in-the-loop oversight. Using a dataset of 962 resumes across 25 categories with an 80/20 stratified split, the system reports standard multi-class performance metrics (classification report and normalized confusion matrix) for the classical pipeline. At the same time, an evaluation harness enables the same artifacts for the LLM baseline on sample subsets. The interface further exposes dataset distribution, top predicted categories, and explanation widgets (SHAP for the classical pipeline; TF-IDF key-term highlighting for the LLM), while applying basic PII filtering. Considerations related to fairness, privacy, and deployment drift are discussed. Comparison with prior SBERT-only systems (e.g., Enhanced Resume Screening using S-BERT) highlights key contributions: (1) a reproducible SBERT→XGBoost→SHAP classification pipeline, (2) an LLM comparator with a standardized evaluation template, and (3) a practitioner-oriented user interface that operationalizes explainability and auditability.
The automotive industry increasingly relies on complex embedded systems that inte-grate software, electronic hardware, electromechanical, and mechanical components. To ensure quality, the A utomotive S oftware P rocess Improvement and Capability Determination (ASPICE) framework has been extended by cybersecurity, which has become a critical concern. Recognizing these risks, the United Nations Economic Commission for Europe (UNECE) introduced Regulation No. 155 in 2021, establishing mandatory provisions for vehicle type approval and cybersecurity management. To address these regulatory demands, ASPICE for Cybersecurity extends the standard ASPICE model with additional processes focused on identifying, analyzing, and mitigating security threats. While the combined assessment of ASPICE and ASPICE for Cybersecurity ensures both process maturity and security compliance, it presents significant challenges for organizations. Conducting separate assessments for ASPICE and ASPICE for Cybersecurity increases time, cost, and organizational workload, as many process areas overlap in scope and evidence. This paper proposes a unified assessment approach designed to evaluate compliance with both ASPICE and ASPICE for Cybersecurity in a single, integrated process. The approach involves systematically mapping equivalent or related process areas across the two models. This approach was applied in an industry case study, demonstrating a reduction in total assessment time and minimizing disruption to development teams. The results indicate that integrated assessments can maintain the rigor and comprehensiveness of separate evaluations while improving efficiency and reducing operational burden.
In July 2024, a failed update in the CrowdStrike Falcon security software disrupted critical services across the United States and several other countries, affecting sectors such as aviation, healthcare, and finance. Despite advancements in automated testing, an error introduced during development was rapidly propagated, exposing vulnerabilities in digital infrastructure and underscoring the risks of technological interdependence. This study examines the incident through an integrative, multivocal literature review that combines academic and gray literature to address three core research questions: the impacts of the outage (RQ1), the technical causes of its severity (RQ2), and the lessons learned for software engineering and cybersecurity (RQ3). The analysis reveals critical failures in validation processes, integration mechanisms, and organizational prac-tices, highlighting the need for robust quality assurance and software governance. The findings contribute to both research and practice by informing more effective strategies for preventing and responding to large-scale cybersecurity incidents.
Automated Program Repair (APR) is currently an essential technique for improving software maintenance, assisting in reducing the burden on developers, and minimizing software defects. Recent advances in large language models (LLMs) have significantly impacted this area by improving bug localization, patch generation, and validation.The current LLM based automated program repair (APR) techniques are systematically reviewed in this paper in order to evaluate their advancements, real world constraints, performance comparisons, and security considerations. The study explores the advantages and disadvantages of open-source alternatives like CodeT5 and UniXcoder as well as proprietary LLMs like GPT4 and Codex, taking into consideration their respective costs, customization, and accuracy. Retrieval augmented generation, finetuning techniques, and hybrid approaches that combine LLM with static analysis tools have all significantly increased the effectiveness of APR. However, problems still persist, such as reliance on AI generated patches, high processing demands, and challenges incorporating these techniques into development processes. Furthermore, ethical issues including biases in created patches and the potential to introduce security defects necessitate further research on validation and security aware patch production. This paper summarizes current research to identify key gaps and proposes future directions to optimize LLM based APR, to assure its reliability and scalability in real world software development.
Online courses have transformed modern education by offering flexible and accessible learning opportunities. However, adapting course content to the diverse learning preferences of students remains a key challenge. This study introduces a data-driven approach to personalize and analyze learning content in online courses based on the Felder-Silverman Learning Styles Model (FSLSM). The concept of “learning styles” is operationalized through four FSLSM dimensions, and “cognitive profiles” refer to learners’ behavioral patterns in processing and interacting with content. An experimental study was conducted with 40 first-year economics students at the University of Guelma (Algeria), divided into control and experimental groups. Both groups accessed the same initial learning materials, while the experimental group received FSLSM-adapted content in the post-test phase. The results suggest a positive effect of the personalized content on learners’ cognitive engagement and progression. This approach demonstrates the potential of integrating learning style analysis and adaptive content delivery in online learning environments.
Context: As software systems become increasingly complex, testing automation plays a critical role in ensuring product quality and reliability. However, the wide variety of available test generation tools with distinct purposes, features, and approaches makes it difficult for practitioners to select the most appropriate option for their projects. Goal: This study aims to identify, analyze, and compare test generation tools reported in both academic and industrial contexts, providing a practical reference guide to support software professionals in tool selection decisions. Method: We conducted a Multivocal Literature Review (MLR) complemented by a two-round survey with 87 software practitioners. The MLR identified tools and features reported in white and grey literature, while the survey assessed practitioners’ familiarity, usage, and perceptions of advantages and challenges associated with these tools. Results: The findings reveal a persistent gap between academic proposals and industrial adoption. While tools such as Postman, Selenium, and Cypress are among the most widely used tools in practice, academic tools like Monkey and Dynodroid remain rarely adopted. Practitioners value visibility, integration, and traceability as key features, whereas inconsistency and maintenance effort are seen as primary challenges. Conclusions: The study contributes a structured reference guide to assist professionals in selecting tools suitable for specific testing contexts. It also provides insights for researchers aiming to align future tool development with industry needs, fostering better usability, integration, and sustainability of automated testing solutions.
Skilled software developers play a crucial role in creating energy-efficient code, contributing to the reduction of carbon emissions and promoting sustainable computing practices. In parallel, Large Language Models (LLMs) have recently emerged as powerful programming assistants, providing support for developers in complex coding tasks. In this context, this study investigates the ability of LLMs to generate computationally efficient implementations of classic algorithms - such as Heap Sort and Binary Search - across different programming languages, with a primary focus on optimizing energy consumption, memory usage, and execution time. Our results show that, in most cases, different LLMs are able to generate these algorithm implementations correctly. Furthermore, in the majority of the evaluated scenarios, LLM-generated code demonstrates superior performance over human-written algorithm code with regard to energy consumption, memory efficiency, and execution time.
Companies around the world spend sometimes many millions for the implementation of ERP-systems (Enterprise Resource Planning systems are used by big organizations and companies for integrated business operation in finance, logistics and human resources) in projects lasting over several years. Nevertheless many of these systems do not deliver the expected benefits. Especially after going live, which is the beginning of the postimplementation phase, in most cases a performance drop can be observed, which lasts for months. There is a special research area in literature known as benefit management", which deals with this problem. Despite comprehensive theoretical concepts benefit management is not common in practice so that the problem of benefits realisation persists even 30 years after the introduction of ERP systems to the market The goal of this thesis is to investigate the benefits realisation for ERP-systems after going life (postimplementation phase) in the real world based on the Mixed Methods Approach. In the qualitative part of the study 44 end users from four industrial companies and ERP consultants were interviewed to find out how benefits materialize, how benefits can be measured and quantified by means of key performance indicators (KPI) and how benefits can be realised over the entire life decle. These interviews were evaluated using the Qualitative Contents Analysis of Mayring. In the quantitative part of the study a set of 12 KPIs (inventory value of raw materials, stock turnover rate of raw materials, cycle time of productions orders, cost deviation of production orders, inventory value of finished goods, stock turnover rate of finished goods, lead time of customer orders, delivery reliability, percentage of customer orders delivered in time, duration of monthly financial closing, number of IT tickets and system usage), which were downloaded as monthly measurement values directly from ERP-systems, was analysed for four participating case study companies over a period of 24 months after going live using SPSS software for Statistical evaluations The result shows that the implementation of an ERP-system is not finished with going live, but requires permanent efforts for benefits realisation to optimize and adjust business processes, provide ongoing end user trainings and further development of an ERP-system in order to exploit the benefit potential of an ERP-system during the whole lifecycle in the best way. The results of the study show the time dependence of the KPIs and that they can be improved by certain measures. Therefor the author proposes a concept for a benefit controlling to support benefits realisation over the entire life cycle of an ERP-system based on a regular evaluation of la set of processual KPIs.
Large Language Models (LLMs) have shown impressive capabilities in natural language understanding and reasoning; however, their internal conceptual organization remains largely opaque. This study introduces a novel prompt engineering approach to explore the latent conceptual spaces of LLMs by identifying maximally divergent conceptual regions. It draws on the conceptual spaces framework, in which knowledge is represented as geometric regions defined by quality dimensions. Within this framework, a proxy metric referred to as the conceptual divergence count is proposed. This metric represents the number of maximally divergent conceptual regions identified through structured prompts. Although the metric does not measure dimensionality in the geometric or architectural sense, it serves as an indicator of a model’s conceptual diversity. The method is applied to a range of LLMs, including models from the Gemini, GPT, and Claude families, as well as additional models such as Llama-3.3-70B-Instruct, Mistral-7B-Instruct-v0.3, Gemma-3-12B-IT, and Grok-3. The results show substantial variation in diversity, with gemini-2.5-flash-preview-05-20 achieving the highest value of 91 and claude-3-7-sonnet-20250219 recording the lowest value of 7. These findings suggest that diversity, as measured by this method, may provide insights into the internal organization of conceptual representations. While the metric is to be an indicator of a model’s factual knowledge structure, providing complementary insight into its internal organization, it does not directly measure knowledge accuracy. The proposed approach contributes a geometric perspective to the evaluation of LLM knowledge, complementing existing benchmarks and supporting interpretability.
As organizations advance their digital transformation efforts, the strategic importance of data quality becomes critical. Legal and security aspects, along with the economic value of data, further emphasize the need for high-quality, reliable, and trustworthy datasets. In particular, the effectiveness of artificial intelligence techniques heavily depends on the integrity of the underlying data. However, organizations often lack structured methods to assess and improve data quality in alignment with business goals. This paper addresses this gap by introducing OKR4DQ, a methodology that leverages Objectives and Key Results (OKRs) to systematically improve data quality. The approach combines quality assessments based on the ISO/IEC 25012 standard with the definition and implementation of OKRs targeting specific quality characteristics requiring enhancement. We present the full methodology and its application in a real-world case study, demonstrating measurable improvements in data quality and offering practical insights into the challenges and benefits of aligning data quality initiatives with business performance objectives.
The growing interest in noninvasive health monitoring has motivated research into using digestive system sounds as diagnostic indicators. This study proposes the Mathematical Formula for Systematic Digestive Sounds (MFSDS), a method that converts abdominal audio signals into mathematical formulas for automated analysis. The approach integrates devices from the Internet of Things (IoT), signal processing, and artificial intelligence to address challenges such as acoustic complexity, environmental noise, and individual variability. Experimental evaluation with datasets of 25, 50, and 100 formulas achieved similarity rates between 50% and 67%, indicating potential for improvement. Although initial accuracy remains below clinical application thresholds, the results suggest that the integration of advanced AI techniques could improve precision and adaptability. The MFSDS framework offers a novel perspective for preventive healthcare and real-time gastrointestinal monitoring.
The article examines predictions made at the end of the 20th century about contemporary society, assessing which have been confirmed or refuted at the start of the 21st century, and explores the impact of the first phase of Artificial Intelligence (AI). Considering contributions from renowned authors such as Bauman, Castells, Kahneman, Nicolelis, and Acemoglu, the study investigates forecasts related to liquid modernity, network society, information era, economic power concentration, and technological advances. The research aims to understand how these predictions influence current society and its policies through a comparative analysis of documents and literature drawing parallels between past forecasts and current developments. Among the findings, Bauman's prediction about volatility and freedom in liquid modernity is confirmed, noticeable in work flexibility and dissatisfaction with consumption. Castells accurately predicted organizational decentralization and global collaboration, despite persistent digital inequality. Rifkin forewarned of job reductions due to automation continuing to face challenges in the current market. Schwab emphasized the Internet of Things (IoT) impact, increasing economic inequality. Chomsky predicted the intensification of corporate power and media manipulation; Nicolelis highlighted the therapeutic applicability of brain-machine interfaces, though limited by ethical issues. Harari emphasized the transformative impact of AI and biotechnology, still holding authoritarian nuances. Acemoglu and Robinson emphasized the importance of inclusive institutions in preventing stagnation. The research highlights the need for ethical governance and regulation of technological innovations, emphasizing the urgency of evolving policies to ensure social equity in the era of technological advancement.
