Background: Virtual reality (VR) integrated with internet of things (IoT) wearable devices offers innovative approaches to mental health interventions by enabling real-time physiological monitoring during immersive therapeutic experiences.Objective: This study aims to evaluate the effectiveness of VR therapeutic games in identifying and measuring emotional responses through physiological signals (heart rate and galvanic skin response) and to classify these responses using machine learning.Methods: We conduct experiments with 103 participants (aged 6-57 years) using FearTherapy, a custom VR game featuring interactions with four animals (Hermit, Bee, Wolf, Spider). Physiological data are collected using Samsung Galaxy Watch 5 for heart rate and Arduino Uno with a galvanic skin response (GSR) sensor. After preprocessing, 55 valid sessions remain for analysis. Individual baseline heart rate values are established and random forest classification with grid search optimization and 10-fold cross-validation is performed.Results: GSR emerges as the most influential feature for classifying emotional states, followed by heart rate difference and baseline reference values. Highest emotional arousal occurs during Spider and Bee interactions. The random forest model achieves 69% cross-validation accuracy and 81% test set accuracy. The model performs well overall but encounters challenges distinguishing Bee from Hermit and Wolf from Spider, suggesting overlapping emotional states.Conclusion: VR therapeutic games combined with IoT physiological monitoring can effectively measure and classify emotional responses. The findings support development of personalized, emotionally adaptive therapeutic interventions for anxiety and phobia treatment, emphasizing the importance of individualized baseline measurements for accurate emotional state assessment.
Background: The vast amounts of data generated by consumers require new forms of processing, in which artificial intelligence stands out for its ability to analyse them more quickly and deeply. However, although there is abundant literature on artificial intelligence (AI) and consumption, most of it focuses on its impact on consumer behaviour rather than its usefulness in enhancing understanding.Objective: The aim of this study is to conduct a thorough review of the existing literature on the use of AI to understand consumer behaviour.Methods: This study uses the PRISMA protocol for the selection of the studies. Then, it combines bibliometric methods with a TCM-ADO framework to review articles. The Scopus database was used to gather peer-reviewed articles from 2014 to 2024. VOS Viewer and R-Studio were utilised for the analysis and visualisation of data.Results: The study provides insights into publication trends, dominant theories, methods, antecedents, decisions and results in the literature about the use of AI to understand consumer behaviour. Furthermore, it identifies potential avenues for future research to advance the development of theory and methodology.Conclusion: Research into the use of AI to understand consumers is still in its infancy. However, everything points to the application of AI in consumer behaviour continuing to expand, and its use for analysing attitudes and behaviour becoming more sophisticated and widespread.
Background: The introduction of retail investors to AI-powered trading platforms and especially on emerging markets, has resulted in a new set of risks linked to algorithmic bias and financial forecasting fairness. Social media sentiment and structured data multimodal strategies have demonstrated a potential, but frequently do not have ethical considerations.Objective: This work proposes a multimodal model predictive control (MPC) framework grounded in fairness-based forecasting of next-day returns on stock in stock market settings, particularly ethical behaviour and transparency of the model on retail markets.Methods: We combine BERT-based sentiment analysis of Reddit discussions and organized stock market indicators and use XGBoost as the fundamental model. Bias is measured using fairness metrics, including demographic parity difference and equal opportunity difference. Debiasing measures such as reweighting and stratified calibration were used to curb the differences in stock categories.Results: The first model has an overall accuracy of 72.3 with the highest accuracy of 83.1 in the case of Tesla - representing bias in the model. Fairness assessment shows some significant differences (DPD=0.23, EOD=0.31), but the mitigation decreases to 0.07. However, the massive performance improvement after adjustment brings up the issue of overfitting or fairness overcorrection.Conclusion: While the proposed debiased framework successfully reduces algorithmic bias, the trade-off between fairness and generalizability underscores the need for caution. These results hold significant implications for digital trading systems and regulatory frameworks of emerging economies such as India, where explainability and fairness of AI models are significant for ethical financial engagement.
Background: Artificial intelligence (AI) is increasingly used both to test software (T1) and to assure AI-based systems (T2), with adjacent software-engineering work that shapes testing practice (T3). Prior reviews are mostly descriptive and rarely report comparable maturity or replicability signals.Objective: To provide a PRISMA-style systematic review (2015-2025, Web of Science) that maps T1-T2-T3 within a testing-centric frame, audits evidence maturity, threats reporting, and artefact openness per paper, and adds an explicit lens of large language models or generative AI (LLMs/GenAI).Methods: We queried the Web of Science Core Collection (2015-2025), screened via a predefined protocol, and extracted ten items (D1-D10) per study to normalize comparisons. Seventy-two papers met the criteria. Findings are organized into three themes: (T1) AI-based software testing, (T2) testing/validation of AI systems, and (T3) AI-related software engineering topics with implications for testing-T3 corresponding to the "beyond" in the paper's title.Results: The corpus is limited in practice-oriented evidence: 31 laboratory/simulation, 3 industrial, 10 hybrid, 6 conceptual/guideline and 22 secondary studies. Only 18/72 provide public artefacts; 33/72 report no empirical metrics. By theme, T1=32, T2=15, T3=25; the LLMs/GenAI subset totals 10 papers. Openness strongly co-occurs with measurable outcomes (88.9% of artefact-sharing papers report metrics vs 42.6% without), yet "all-three credible" studies (industrial/hybrid + open artefacts + metrics) are rare (4/72 overall; 1/10 for LLMs/GenAI).Conclusion: AI shows promise for testing, but evidence remains thin on industrial adoption and reproducibility. We recommend prioritizing hybrid/industrial validations, releasing artefacts by default, and using standardized task-metric bundles. The review presents T1 and T2 results, separates T3 for scope clarity, and provides actionable maturity and replicability signals to guide responsible, empirical adoption.
Background: Large Language Models (LLMs) have transformed research and industry applications; however, cloud deployment decisions remain complex and poorly documented, particularly for academic researchers operating under budget constraints. Systematic guidance on infrastructure selection for LLM-based research is limited.Objective: This study provides a comprehensive empirical evaluation of cloud-based LLM deployment architectures, examining inference efficiency, serverless platform availability, and architectural trade-offs across major cloud providers to deliver actionable guidance for budget-constrained researchers.Methods: The author evaluated 32 open-source LLMs ranging from 0.6 billion to 1 trillion parameters across serverless and Bring Your Own Container (BYOC) deployment configurations. Using the Belebele benchmark, we analyzed cost-efficiency relationships, serverless platform availability, and metrics exposure across Amazon SageMaker, Amazon Bedrock, Azure Serverless, and Hugging Face-compatible providers.Results: Model performance follows a logarithmic scaling relationship with parameter count (R2=0.727) and deployment cost (R2=0.639). Models in the 30-50B parameter range achieve 85-90% of maximum accuracy at a fraction of the cost of frontier models. However, serverless availability remains fragmented: only 34.4% of examined models are accessible via serverless endpoints, with minimal cross-platform redundancy (6.2%). Deployment architecture introduces a fundamental trade-off: serverless platforms expose 71% fewer metrics than BYOC approaches while eliminating infrastructure management overhead and idle costs.Conclusion: These findings provide practical guidance for researchers selecting cloud infrastructure under budget constraints. Models in the 7-14B range offer optimal cost efficiency, while the 30-50B range maximizes accuracy per dollar for demanding tasks. The results also challenge the prevailing emphasis on ever-larger models, as diminishing returns become substantial beyond 30B parameters. Persistent gaps in serverless availability and observability highlight the need for greater standardization in cloud platforms.
Background: The rapid development of high-speed railway (HSR) systems requires advanced and reliable communication infrastructure to support operational safety, passenger services and intelligent transportation functions. Despite growing attention, comprehensive reviews of scientific developments in HSR communication systems remain limited.Objective: This article seeks to explore the development, thematic landscape and prospective directions of HSR communication studies through bibliometric analysis, emphasizing the identification of core technologies, prevailing research trends and promising avenues for future investigation.Methods: A total of 352 articles published between 2005 and 2024 were retrieved from the Scopus database to examine research development in the field. The dataset was cleaned using OpenRefine and analysed through keyword co-occurrence techniques in VOSviewer and Biblioshiny. This analysis identified thematic clusters, temporal trends and core technologies shaping the research domain.Results: The study highlights four major research clusters: artificial intelligence (AI) and adaptive communication technologies, fifth-generation (5G) network architecture and mobility solutions, signal processing and quality of service (QoS) optimization and channel modelling with propagation characteristics. Massive multiple-input multiple-output (massive MIMO) and millimetre-wave (mmWave) technologies emerge as key enablers for addressing high-mobility challenges. Furthermore, the findings reveal a growing integration of AI, edge computing and real-time communication protocols in recent research.Conclusion: This overview offers a macro-level perspective on the scientific landscape of HSR communication studies. The findings underscore the growing adoption of adaptive, intelligent and energy-efficient technologies, providing strategic guidance for future scholarly work and policymaking in advancing next-generation railway communication systems.
Background: Some of the world's most valuable platform businesses rely on products and services provided by small and medium-sized enterprises (SMEs). Though, the modern digital platform economy is increasingly shaped by uncertainties and power asymmetries benefitting dominant platform owners and threatening smaller players participating as complementors in those ecosystems. Negative consequences include lock-in effects and platform dependency, exploitative participation terms and eroded entrepreneurial autonomy on the SMEs' side, which altogether harm the digital platforms' long-term viability, too. Addressing these issues, this paper investigates design principles for digital platforms taking SME complementors' needs into account.Objective: This study investigates design principles for digital platforms that enhance suitability for SMEs as complementors, focusing on stakeholder-centric platform design approaches that better accommodate SME-specific needs and requirements. In this, this study aims to address current negative developments concerning imbalanced power dynamics and uncertainties emerging from platform owner-SME-partnerships in dominant digital platform ecosystems.Methods: A qualitative reflective meta-analysis approach was employed, combining explorative expert interviews with SME specialists and a thematic literature review of SME platform design research. The methodology synthesized findings to identify meta-requirements and derive design principles through interpretive analysis.Results: Eleven meta-requirements were identified and synthesized into four design principles: the principle of SME empowerment, the principle of open boundaries, the principle of transparent and fair participation terms, and the principle of reflection of individuality. In combination, these principles aim to inform future digital platform design that takes SME complementor needs into account.Conclusion: The conceptual proposal of four design principles guides researchers and practitioners in creating SME-suitable digital platforms with stakeholder-centric design approaches. The principles enable digital platform models that accommodate diverse SME requirements, enhance participation experiences, and foster collaborative ecosystems tailored to SME characteristics and their operational contexts.
Background: SHapley Additive exPlanations (SHAP) methods are widely used to interpret machine learning models, yet most implementations assume feature independence. This assumption rarely holds in practice, especially when features are correlated, leading to biased and unstable attributions.Objective: We introduce Corr-SHAP, a correlation-aware SHAP approach that produces more faithful and stable feature attributions by explicitly modeling feature dependencies. Our aim is to enhance the accuracy, robustness, and scalability of SHAP explanations for models trained on correlated data.Methods: Corr-SHAP models feature correlations via a multivariate Gaussian approximation with a Ledoit-Wolf covariance estimator. We design a correlation-aware sampling distribution that penalizes redundant coalitions, improving computational efficiency in higher dimensions. To correct the induced bias, we employ a Self-Normalized Importance Sampling estimator, which re-weights samples by the ratio of the true Shapley kernel to the sampling probability. Our analysis establishes high probability error bounds in terms of Effective Sample Size, extending convergence guarantees to correlated feature spaces.Results: Across synthetic and real-world datasets, Corr-SHAP achieves Shapley value estimates that closely align with Kernel SHAP, while exhibiting substantially lower variance and more stable feature rankings. In correlated clusters, Corr-SHAP systematically down-weights redundant features, improving ranking fidelity without introducing bias. To further support scalability, we demonstrate that combining Corr-SHAP with Leverage-SHAP reduces variance in higher-dimensional settings.Conclusion: Corr-SHAP provides a statistically grounded and computationally efficient framework for SHAP value estimation under feature correlation. By integrating correlation modeling, bias correction, and variance reduction, it scales beyond small toy problems and delivers explanations that are both accurate and reliable, making it a valuable tool for practitioners analyzing complex real-world datasets.
Background: The growth of high-throughput sequencing and multi-omics research has intensified the need for secure, interoperable and transparent data management infrastructures. Blockchain technology has been widely proposed as a potential solution; however, its feasibility, empirical maturity and comparative performance in bioinformatics remain unclear.Objective: This systematic review analyses blockchain applications in bioinformatics, highlighting claimed security and governance benefits, comparing them with traditional data security approaches, discussing implementation challenges and assessing the empirical rigor of existing studies using a structured quality assessment framework.Methods: This overview was conducted using Scopus, ScienceDirect, IEEE Xplore, ACM Digital Library and SpringerLink for publications from 2014 to 2024. Search strings combined blockchain, bioinformatics and security-related terms. Sixty-five studies met the inclusion criteria. Each study was evaluated using five equally weighted quality dimensions: application specificity, clarity of benefits, empirical evaluation, challenge articulation and reproducibility.Results: Most studies focused on blockchain use cases in genomic data sharing, provenance tracking and access control, with a strong emphasis on conceptual benefits such as immutability and auditability. Fewer studies provided empirical evaluations or direct comparisons with traditional security mechanisms. Quality assessment results revealed a predominance of conceptual and prototype-level contributions; over half of the studies lacked empirical benchmarking and reproducibility was frequently limited. Heterogeneity in blockchain architectures and the absence of standardized genomic benchmarking environments hindered cross-study comparison. No study demonstrated deployment within a production-scale genomic pipeline.Conclusion: Blockchain demonstrates conceptual potential for enhancing provenance, decentralized governance and tamper-resistant auditing in bioinformatics data management. However, empirical validation remains limited and significant technical, regulatory and organizational challenges persist. The current evidence base is insufficient to support large-scale adoption. Future research should prioritize benchmarking using realistic genomic workloads, hybrid architectures that integrate off-chain storage, consent-aware governance models and alignment with regulatory frameworks such as GDPR and HIPAA.
Background: The integration of artificial intelligence (AI) in healthcare depends on striking a balance between patient privacy and clinical utility. The standard methods often compromise one for the other, preventing the development of trustworthy healthcare AI.Objective: This paper aims to resolve the privacy-utility trade-off by developing an enhanced federated learning framework with adaptive differential privacy (DP) mechanisms that are optimized for clinical data.Methods: We implement and compare several different methods, from the most centralized deep learning to various federated configurations with formal DP guarantees. Our improved framework involves adaptive noise scheduling and quality-weighted federated averaging on top of a federated neural network framework. We validate on two major diabetes screening datasets: Diabetes Health Indicators (BRFSS 2015) and National Health and Nutrition Examination Survey (NHANES 2015-2016), including comprehensive clinical measurements.Results: This paper presents a favourable balance between privacy protection and clinical utility for both datasets. It offers strong formal differential privacy guarantees and good diagnostic performance, achieving high ranking accuracy with clinical risk prioritization. The model demonstrates generalization robustness by capturing clinically meaningful risk factors aligned with established medical guidelines, confirming that the applied privacy-preserving mechanisms do not compromise clinical relevance.Conclusion: Our framework meaningfully advances the privacy-utility trade-off healthcare AI, by offering tunable formal privacy guarantees while ensuring strong clinical performance. The approach is highly generalizable across diverse data collection methodologies and maintains clinically relevant feature representations, thus allowing safe adoption in sensitive medical domains.
Background: Artificial intelligence (AI) has become a fundamental part of everyday life, making it crucial to integrate AI into the information society in ways that protect individual rights.Objective: This study explores the perspectives of different stakeholders on the ethical use of AI. The aim of this research is to identify practical measures that can help address ethical challenges associated with AI deployment.Methods: A scoping literature review approach was adopted, focusing on the most relevant articles addressing the ethical aspects of AI usage from Web of Science Core Collection and Scopus databases. The analysis was performed with focus on the perspectives of four key stakeholders: policymakers, AI innovators, business leaders, and individuals.Results: Findings highlight key measures to promote ethical AI usage: technical, organisational, regulatory, and individual measures. In this context: (1) policymakers are responsible for establishing governance and regulations; (2) AI innovators must embed ethics into AI systems; (3) business leaders should establish ethical policies and guidelines; and (4) individuals need to think critically and use AI responsibly.Conclusion: The responsible deployment of AI requires a comprehensive approach that involves the collaboration of all relevant stakeholders. The future development of AI relies on the adoption of ethical guidelines and the assurance of responsible AI system design.
Background: To objectively evaluate the capabilities of large language models (LLMs), we need to develop toolsthat enable such assessment. While numerous benchmarks exist, the vast majority are in English and focus on general knowledge, often overlooking the cultural and factual specifics of smaller countries. Objective: Currently, there is no benchmark that tests LLMs' knowledge of Slovak realia. At the same time, LLM performance in this domain remains inadequate. To objectively measure and compare these capabilities, our goal is to develop and validate a specialized benchmark for assessing LLMs' knowledge of Slovak cultural and factual context. Methods: We created a set of 35 questions on Slovak culture, geography, history and language. We designed them to provide unambiguous answers suitable for automated evaluation. Subsequently, we presented the questions to three major language models-DeepSeek V3, OpenAI GPT-4o and Llama 3. Results: DeepSeek scored 54% of correct answers, OpenAI GPT scored 51% and Llama scored 40%. The models scored best in geography questions. Overall scores showthat models are not verygood in recognising Slovak realia. Conclusion: We present the benchmark for evaluating large language models on Slovak-related knowledge. Even the most advanced current models, including OpenAI GPT and DeepSeek, answered only around half of the questions correctly. This highlights a significant gap in international LLMs' understanding of culturally specific facts, underscoring the need for specialized, nationally tailored language models.
Background: System Development Life Cycles (SDLC) are proposed through development methodologies (e.g. Rational Unified Process) and international standards (e.g. ISO/IEC 12207) to guide the systematic development of software products to meet expected time, budget, and functional quality. The ISO/IEC 29110-5-1-2: Software engineering guidelines for the generic Basic profile standard provides a disciplined-systematic lightweight SDLC alternative to agile approaches for businesses interested in ISO/IEC certifications rather than agile ones. However, its utilization in the domain of Big Data Analytics Systems has not been investigated.Objective: Big Data Analytics Systems (BDAS) have been proposed using rigorous SDLCs such as CRISP-DM, Team Data Science Process, and Domino Data Science Lifecycle, but their utilization for small business - or small teams called Very Small Entities (VSEs) - is scarcely reported given the requirements demanding human, technological and financial resources not available in small business. Given this problematic situation, new agile SDLCs for BDAS have been proposed such as Data-Driven Scrum, but some small organizations still require the utilization of a more systematic development process, and thus SDLCs based on ISO/IEC standards are expected.Methods: In this research, the design of Light Data Science - Analytics Methodology (LDSAM) and its pilot usability evaluation from a sample of 27 international practitioners are reported. LDSAM is a lightweight development methodology aligned to the ISO/IEC 29110 - Basic Profile - created especially for VSEs.Results: LDSAM was elaborated using Design Science Research Methodology. Initial usability evaluation results of LDSAM are satisfactory, but further empirical research is encouraged to advance to more mature and stable lightweight SDLCs for BDAS.Conclusion: The LDSAM was successfully developed, and its usability was evaluated by a pilot sample of international academics and professionals. Favorable results were obtained on the usability metrics for the proposed LDSAM SDLC for BDAS.
Background: Environmental monitoring and data collection critically depend on wireless sensor networks (WSN). However, the limited battery life of these sensor nodes significantly affects the operational lifetime of the network. This often results in challenges related to energy efficiency and clustering tasks due to dynamic network topologies and limited resources.Objective: This paper aims to improve clustering efficiency and extend the network lifetime of WSN by proposing a hybrid model.Methods: The proposed framework was evaluated in terms of energy consumption, network lifetime, Packet Delivery Ratio (PDR), and clustering efficiency through extensive simulations. Modified Salp Swarm Algorithm (MSSA)-Deep Stacked Sparse Autoencoder (DSSA)-K-means is a promising solution for next-generation WSN in smart environments and industrial Internet of Things (IoT) applications because experimental results show that it achieves higher clustering accuracy, energy economy, and robustness against network failures than current methods.Results: Extensive simulations were conducted to evaluate the proposed framework based on energy consumption, network lifetime, PDR, and clustering efficiency. The experimental results demonstrate that the proposed MSSA-DSSA-K-means model achieves higher clustering accuracy, energy economy, and robustness against network failures compared to existing methods.Conclusion: The paper concludes that this hybrid model is a promising solution for next-generation WSN in smart environments and industrial IoT applications.
Background: The sharing of electronic health records among hospitals is crucial for ensuring consistent patient treatment. However, the process remains challenging due to the existence of varied systems, privacy concerns and interoperability issues. It is often difficult to maintain an equilibrium of security, efficiency and compliance across all platforms.Objective: The objective of this article is to develop a framework that enables secure, efficient and interoperable Electronic health records (EHR) sharing across healthcare systems.Methods: The proposed work introduces a Lightweight elliptic curve cryptography proxy re-encryption (LWECC-PRE) framework that facilitates safe and distributed EHR exchange through Ethereum and Hyperledger Fabric blockchains. It integrates Hybrid elliptic curve proxy re-encryption (HEC-PRE) by combining elliptic curve cryptography with proxy re-encryption for giving healthcare providers the means to control access to confidential data. Besides, the design incorporates the Elliptic curve integrated encryption scheme (ECIES) for secure data encryption and the Elliptic curve digital signature algorithm (ECDSA) to verify the integrity and authenticity of data communications. It uses Interplanetary file system (IPFS) for secure peer-to-peer storage. The design supports asynchronous record sharing between blockchain networks through smart contracts and regulated re-encryption.Results: The experimental results show that the framework minimizes computation overheads, preserves patient privacy and improves interoperability of distributed healthcare systems.Conclusion: The proposed solution addresses key challenges in EHR sharing by providing a safe, secure, efficient and patient-centred solution to healthcare data exchange.
Background: The rapid evolution of software engineering has positioned DevOps practices and Continuous Testing (CT) as critical approaches for achieving speed, quality, and reliability in software delivery. Test automation is central to CT, yet its adoption remains inconsistent due to a complex interplay of technological, organizational, and environmental conditions.Objective: This study employs a systematic literature review guided by the Technology-Organization-Environment (TOE) framework to identify, categorize, and synthesize the antecedents that influence the test automation adoption in DevOps continuous testing.Methods: Using the PRISMA protocol, 49 peer-reviewed studies published between 2015 and 2025 were systematically analyzed, yielding 61 distinct factors comprising 29 technological, 19 organizational, and 13 environmental antecedents. These factors were further consolidated into thematic clusters to enhance analytical clarity and reduce fragmentation.Results: The findings demonstrate that technological and organizational drivers, including relative advantage, compatibility, top management support, and employee competence, dominate the literature, while environmental influences such as competitive pressure, regulatory requirements, and vendor ecosystems are comparatively underexplored. This imbalance indicates that although the TOE framework is widely applied in technology adoption studies, empirical research has given greater attention to internal adoption enablers than to external pressures. By simplifying and synthesizing the factors into coherent sub-themes, this study contributes to both theory and practice by offering a structured lens through which test automation adoption can be examined in the DevOps CT context. Theoretically, it validates the relevance of TOE for analysing multidimensional adoption dynamics, while practically, it provides managers with evidence-based insights to prioritize critical factors when planning automation initiatives. Methodologically, it demonstrates the importance of transparent and replicable review processes for advancing cumulative knowledge.Conclusion: Overall, the study bridges fragmented findings into a coherent framework and strengthens understanding of adoption strategies in continuous testing environments.
Background: Osteoporosis is a condition characterized by bones that are porous and brittle, increasing the risk of fractures. It is often asymptomatic until substantial harm develops, making it crucial to treat at the onset of the disorder.Objective: This study aims to develop a practical, in-depth and adaptable framework utilizing constructed clinical and demographic datasets for the early detection of osteoporosis.Methods: We design a cascade convolutional neural network with adaptive weight fusion and fine-tune it with a real-coded genetic algorithm. An anonymized clinical and demographic record publicly available bone mineral density dataset was used. Missing data identification, normalization and encoding of categorical variables were key diagnostic steps. The training subset consisted of 70% of the dataset, while the remaining 30% was used for testing.Results: The predictive capability of the proposed model is demonstrated by utilizing two datasets. Dataset 1 is used for training and testing, achieving a classification accuracy of 99.5%, precision of 98.7%, recall of 99.0% and AUC-ROC of 0.99. Dataset 2 is used to test the model generalizability, achieving a classification accuracy of 97.0%.Conclusion: The model integrates well into primary care settings, as it relies on structured clinical data rather than imaging. Its low costs relative to value and high scalability make it suitable for population-level screening and treatment of osteoporosis. The limitation of the research is that we utilize only a clinical dataset to train the model without image analysis.
