Data and Metadata (Journal)
Not a member yet
727 research outputs found
Sort by
AI-Driven Hybrid Defense Mechanisms for Enhancing Cybersecurity in Cyber-Physical Systems Through Packet Sniffing and Cyber Ranges
Full text linkIntroduction: Cyber-Physical Systems are the backbone of modern critical infrastructures but remain inherently vulnerable to cyberattacks due to their interconnected nature. This calls for more adaptive and intelligent intrusion detection solutions, as existing approaches often fall short in capturing the spatial-temporal complexity of CPS traffic.Methods: This work proposes a hybrid deep learning framework based on the integration of CNN and LSTM networks with an attention mechanism. The system exploits real-time packet sniffing for fine-grained traffic analysis and the use of cyber range simulations to evaluate its performance in different attack conditions. A structured preprocessing pipeline, covering normalization, time windowing, and controlled data augmentation, ensures high-quality feature extraction while maintaining spatial and temporal patterns.Results: The proposed model outperforms standalone CNN and LSTM architectures on a balanced multi-class CPS dataset with 99,08 % accuracy and very high precision, recall, and F1-scores across all attack types. Attention significantly enhances sensitivity by picking up important temporal features and provides better interpretability via packet-level relevance mapping. The model maintains an extremely low false-positive rate, further supporting its suitability for real-world deployment.Conclusions: These results position the hybrid CNN-LSTM-Attention architecture, combined with packet sniffing, as a robust and adaptive intrusion detection for CPS environments. Strong performance with low error rates accordingly underlines the potential to mitigate emerging threats. Future work will extend the evaluation to diverse datasets and will benchmark the system against state-of-the-art detection models in order to further validate generalizability
Analyzing the Impact of Self-Regulated Learning on Academic Performance and Student Engagement
Full text linkSelf-Regulated Learning (SRL) refers to students’ ability to manage and control their own learning processes, and it has been widely recognized as a key predictor of academic achievement and student engagement. Despite the growing volume of SRL research, a Scopus-based analysis indicates that studies specifically examining the relationship between SRL, achievement, and engagement remain limited. This gap restricts the development of more comprehensive educational strategies that maximize student learning outcomes.This study aims to analyze trends in SRL research within the 2020–2024 period, particularly focusing on how SRL relates to academic achievement and student engagement, and to identify underexplored themes that require further investigation.Results: The analysis reveals that although SRL has been extensively studied, research explicitly linking SRL to academic achievement and student engagement remains scarce. Network visualization shows that Cluster 5 contains the least-developed themes, including achievement goals, learning strategies, metacognition, and motivation indicating that these topics require deeper exploration. Overlay visualization further demonstrates that the peak of SRL-related research occurred in mid-2021 and throughout 2022, followed by a noticeable decline in 2023 and 2024. Density visualization confirms that while SRL is a widely researched construct, studies investigating its direct impact on achievement and engagement are still relatively limited
Presence of Ecuador in the Web of Science from open access in post-pandemic period 2019- 2021: A multivariate analysis
Full text linkIntroduction: The COVID-19 pandemic has profoundly transformed scholarly communication, accelerating the global adoption of open access (OA). In this context, it is relevant to analyze how Ecuador positioned itself in these dynamics in the post-pandemic period.Objective: To evaluate Ecuador’s presence in the Web of Science (WoS) database from 2019 to 2021 and identify the impact of the pandemic on scientific production and the adoption of open-access models.Method: A total of 9085 articles indexed in WoS under the affiliation “Ecuador” were retrieved. Data analysis was performed in R using multivariate statistical techniques and visualization tools (HJ-Biplot), complemented by Bonferroni tests at the 95 % confidence level to compare citation differences between OA and subscription publications.Results: Of the total publications, 52 % corresponded to open access. These articles received more citations on average than subscription-based articles, with statistically significant differences. Private universities accounted for 43 % of publications, public universities for 42 %, and collaborative works for 15 %. A progressive shift toward OA was evident, especially after 2020, with the green route predominating over the gold, bronze, and hybrid pathways.Conclusions: Ecuador has notably transitioned toward open access, enhancing the visibility and impact of its scientific production. However, challenges remain, related to the lack of national policies and limited inter-institutional collaboration. Strengthening OA strategies is recommended to democratize knowledge and improve the international positioning of Ecuador’s scientific output
Relationship between absorptive capacity and innovation capacity in firms: An empirical analysis using PLS-SEM
Full text linkSmall and medium-sized enterprises (SMEs) constitute a fundamental part of any country\u27s economic structure, representing self-employment initiatives that generate jobs and contribute significantly to the Gross Domestic Product (GDP). However, due to the high level of competition they face, they experience limited growth and a low adoption rate of innovations, which in many cases leads to their short-term demise. Hence the importance of their capacity to absorb external knowledge, which allows them to adapt to changes in the environment and create innovations that enable them to compete in today\u27s turbulent economic and financial market. In this context, the present study focuses on analyzing the relationship between the absorptive capacity and the innovation capacity of SMEs in the state of Sinaloa, Mexico. The method applied was PLS-SEM structural equation modeling, with a statistical sample of 151 companies. The results suggest that SMEs prioritize the assimilation and transformation of knowledge to foster innovation, particularly in their innovative performance, innovation culture, process innovation, and service innovation. The study\u27s contributions include the theory of dynamic capabilities for organizations, as well as sufficient empirical evidence to demonstrate the correlation between absorption and innovation that streamlines the design of business strategies
Development of a new predictive hiring system with multi-model voting sets and advanced stacking techniques for assessing semantic soft skills
Full text linkHuman resources face a major challenge in extracting and identifying the semantic correspondence of data, in particular the soft skills most recruiters seek from heterogeneous data. The complexity lies in identifying the relationships between textual descriptions in CVs, keywords, descriptions in professional networks, and relevant soft skills such as communication, persuasion skills, negotiation, relationship building, empathy, teamwork, conflict resolution, emotional intelligence, time management, work ethics, after analysis and research, we chose these soft skills as input data because they encompass all the soft skills that a recruiter might look for in a candidate for any position. The present study introduces a predictive hiring system that assesses candidate performance based on soft skills extracted from three main sources, namely resumes, professional social network profiles, and psychometric assessments. A dataset of over one million candidate records was processed. Data analysis relied on state-of-the-art NLP techniques, including word embeddings and contextual language models, in order to build a semantic database linking keywords, phrases, and descriptions to targeted soft skills. Machine and deep learning models were applied, followed by an ensemble approach integrating KNN, Decision Tree, and Random Forest. To overcome prediction accuracy and overfitting limitations, a meta-model XGBoost was developed, achieving superior results with an accuracy of 98%. The results demonstrate that the proposed meta-model outperforms baseline approaches, delivering high predictive accuracy and robust generalization. These findings highlight the potential of combining semantic analysis with advanced machine learning to support more reliable and scalable predictive recruitment systems
Artificial Intelligence in Higher Education 5.0: Ethical Implications, Pedagogical Innovation and Personalized Learning
Full text linkIntroduction: the incorporation of artificial intelligence (AI) into Higher Education 5.0 transformed pedagogical models and institutional ethical frameworks, consolidating AI as a key driver connecting technological innovation and human-centered education. However, scientific literature revealed conceptual fragmentation that limited a comprehensive understanding of its ethical, pedagogical, and personalization impacts.Method: a narrative integrative review was conducted between January and September 2024 using the Scopus, Web of Science, SciELO, ERIC, and Redalyc databases. Inclusion criteria focused on indexed publications from 2019–2024 with verifiable DOIs addressing ethics, pedagogical innovation, and personalized learning. Out of 146 identified documents, 32 studies were selected through coding, thematic comparison, and theoretical triangulation.Results: findings showed that 41% of the studies focused on ethical implications, 34% on pedagogical innovation, and 25% on personalized learning. Most publications originated from Scopus Q1 and Q2 journals. Results evidenced a trend toward hybrid, student-centered ecosystems, increased use of learning analytics, and the need for robust institutional ethical frameworks.Conclusions: aI was consolidated as a strategic driver for Higher Education 5.0, capable of fostering inclusion, equity, and teaching transformation. Nevertheless, gaps persisted in digital governance, ethical training, and critical evaluation of technological impact.
IoT and AI for Smart Rural Fishing: System Architecture, TEK Integration, and Economic Viability Analysis
Full text linkIntroduction: The integration of IoT and AI technologies presents opportunities for modernizing traditional fishing practices while preserving ecological knowledge.Objective: This research develops and evaluates an integrated IoT-AI system to enhance traditional Chinese fishing net operations in Kerala backwaters, India, with an emphasis on TEK integration and economic viability assessment.Method: We deployed an edge-cloud computing architecture integrating 15 environmental sensors, automated winch systems, and cloud-based AI analytics at Chathedam fishing site (9,9674°N, 76,2816°E) over six months (January–June 2025), documenting 1,000 fishing operations and validating 20 TEK rules through statistical analysis.Results: Nine high-confidence TEK rules (≥0,80) achieved 100 % validation success with 18–47 % catch improvements. AI-guided operations achieved 10 percentage point improvement in profit margin (70,4 % vs 60,4 %) primarily through cost reduction. System investment of $1,720 achieves payback in 1,8 months with 898 % ROI over 18 months.Conclusions: The validated edge-cloud architecture, TEK-AI integration framework, and demonstrated economic viability provide a replicable model for technology-enabled enhancement of traditional small-scale fisheries
Improving Autism Detection Accuracy with an Optimized Local-Asymmetric Adaptive Hybrid GCN for EEG Data
Full text linkAutism spectrum disorder (ASD) is an intricate nervous disorder typically diagnosed through the use of electroencephalography (EEG). A novel model named Dual Encoder-Balanced Conditional Wasserstein Generative Adversarial Network with Resting-state EEG-based Hybrid Graph Convolutional Network (DEBCWGAN-Rest-HGCN) was made from this context. By fixing the class imbalance and making synthetic EEG samples, it was able to detect ASD with encouraging results. However, it ignores the dynamic brain patterns recorded by task-based EEG in favor of resting-state EEG. The Rest-HGCN model also cannot successfully capture the uneven spatial and temporal aspects of EEG signals, and its fixed hyperparameters might make it less accurate in detecting different types of EEG data. This article presents a new model for finding and diagnosing ASD called the Optimized Local-Asymmetric Adaptive Hybrid GCN (OLA2HGCN). This model uses both spatial and temporal information from resting-state and task-driven EEG signals. It is based on the way autism affects brain connections and a variation amid the left and right hemispheres. The LA2HGCN can efficiently collect discrete spatiotemporal EEG information through distinct areas and hemispheres by improving the HGCN model with hierarchical feature extraction and fusion approaches. This model has a time based feature extraction approach in the cognitive prior graph branch that picks up temporal characteristics inside and between brain areas. It also has an adaptive GCN for spatial feature extraction across non-Euclidean distributions of electrodes. An attention layer shows how each hemisphere helps with ASD classification. A new Quantum Artificial Gorilla Troops Optimizer (QGTO) is also presented to help the LA2HGCN model choose the best hyperparameters. The QGTO is based on the social intelligence of gorilla tribes. It rapidly traverses intricate search spaces and achieves an equilibrium between exploration and exploitation. By adding quantum mechanics to the GTO method, it can better find its way through complicated search spaces and stay away from local optima. This makes hyperparameter selection more successful. Finally, the test results show that the DEBCWGAN- OLA2HGCN on the EEG Dataset for ASD and the ABC-CT dataset are 95.04% and 92.27% accurate, respectively, when compared to other algorithms
Mathematical model based on Machine Learning Enabled IoT Sensing and Decision-Making in Farming
Full text linkIntroduction: The rapid expansion of global food demand, combined with unpredictable climate variability and resource scarcity, necessitates intelligent solutions for sustainable agriculture. Method: This study introduces an IoT-driven intelligent greenhouse monitoring and decision-making framework that integrates advanced machine learning (ML) models with heterogeneous environmental data. Using multi-source sensor networks and edge-cloud collaboration, the framework dynamically regulates greenhouse environments while providing yield forecasting and disease detection capabilities. Results: Experimental results demonstrate that the proposed system achieves high detection accuracy (F1 = 96.8%), low yield prediction error (RMSE = 0.40 tons/ha), and efficient energy usage (0.46 J per inference). Reinforcement learning controllers further optimize climate regulation, reducing temperature RMSE to 0.72 °C and achieving energy savings of up to 20% compared to traditional PID systems. The hybrid CNN-Transformer disease detection model outperforms benchmarks, attaining 97.9% accuracy with improved calibration reliability. Conclusion: Collectively, these findings confirm that the proposed IoT–ML framework not only improves productivity and sustainability but also ensures scalability for large-scale deployments in diverse agricultural environments
Deep Spatiotemporal Analysis of Cardiac Motion from Video and Range Data Images for Early Detection of Heart Diseases
Full text linkIntroduction; The detection of motion-based cardiac abnormalities at an early stage proves difficult because individual systems fail to measure how motion affects depth and structural changes. A multimodal spatiotemporal system would enhance the accuracy of medical diagnoses. Objective; The research aims to create a real-time system which unites cardiac video data with range/depth information to detect cardiac conditions at an early stage. Method; The system operates through independent encoders which join their data streams through a gated fusion module. The system performs denoising operations followed by statistical normalization and geometric transformation of the input data. The system uses beat-level temporal attention to identify essential time segments for clinical evaluation. The research evaluated system performance through comparison with video transformers and traditional temporal analysis methods. Result; The model produced F1 reached 0.945 while AUROC reached 0.9978 and the model achieved sensitivity at 0.950 and specificity at 0.940 and precision at 0.940 and AUPRC at 0.972. The system demonstrated excellent calibration performance through its ECE and Brier values which approached perfect results (slope ≈ 1.01, ≈ 0). The system produced useful screening results when using 10% and 20% thresholds which produced 0.142 and 0.118 respectively. The system performed real-time processing at 4.9 GFLOPs while maintaining a processing time of ~98 ms. Conclusion; The combination of intensity dynamics with depth-derived geometry allows for accurate real-time cardiac prediction with precise calibration. The proposed method delivers superior results than single-signal systems and conventional temporal methods which makes it a useful advancement for early detection and point-of-care cardiology