TULTECH Journals
Not a member yet
235 research outputs found
Sort by
A Unified AI Framework for Confidentiality Preserving Cyberattack Detection in Healthcare Cyber Physical Networks
No full textHealthcare Cyber-Physical Systems (HCPS) are increasingly exposed to sophisticated cyberattacks that compromise both service continuity and patient privacy. Existing intrusion detection systems (IDS) based on federated learning (FL) and differential privacy (DP) have demonstrated potential, but most lack adaptive privacy controls and hybrid learning strategies for detecting zero-day threats. This study proposes an innovative unified IDS framework that integrates (i) a hybrid machine learning fusion of supervised (SVM, RF), unsupervised (Autoencoder, Isolation Forest), and ensemble methods to improve both known and unseen attack detection, and (ii) an adaptive DP noise control mechanism, which dynamically adjusts privacy levels during federated aggregation to optimize the privacy–utility trade-off. Experiments were conducted using the Healthcare Intrusion Detection Benchmark Dataset long with validation on supplementary healthcare IoT traces, enabling reproducibility and robustness testing. Results show that the proposed framework achieves 97.5% accuracy, 96.8% precision, 95.9% recall, F1-score of 96.3%, and AUC-ROC of 98.2% without DP, and maintains competitive performance at strict privacy settings (ε=0.1) with 85.3% accuracy and F1-score of 83.4%. Comparative analysis against baseline IDS models (SVM, CART, CNN) and state-of-the-art privacy-preserving IDS frameworks confirms the superiority of the proposed system in zero-day attack detection, scalability, and HCPS-specific applicability. The findings demonstrate that adaptive, privacy-preserving IDS solutions are feasible for real-world digital healthcare environments
Unveiling Anomalies: Leveraging Machine Learning for Internal User Behaviour Analysis – Top 10 Use Cases
No full textInsider threats pose a significant risk to organizations, as traditional Security Information and Event Management (SIEM) systems struggle to detect subtle, evolving anomalies in user behaviour. While machine learning (ML) offers promise, the absence of a structured approach to prioritize and validate high-impact threat scenarios limits its practical adoption. This research addresses this gap by systematically identifying and validating the top 10 critical insider threat use cases—including data exfiltration, privilege escalation, and lateral movement—through a methodology combining MITRE ATT&CK tactics, Verizon Data Breach Investigations Report (DBIR) statistics, and related research papers. We then integrate the Random Cut Forest (RCF) algorithm into the Wazuh/OpenSearch SIEM platform, tailoring its unsupervised learning capabilities to detect these prioritized threats in real time. By correlating ML-driven anomaly scores with rule-based alerts, our solution reduces false positives by 35% and achieves a 94% true positive rate for high-risk use cases like unauthorized access. Validation in a production environment confirms the framework’s efficacy, with detection times under 3 minutes for 80% of anomalies. Beyond technical integration, this work establishes a replicable blueprint for aligning ML models with operational priorities, empowering organizations to focus resources on the most damaging insider threats
3D Magnetic Resonance Image Segmentation Using HD Brain Extraction in 3D Slicer
No full textApplications of image processing in radiology and radiation are critical for the development of models, simulations, and computational tools. 3D Slicer is a widely used platform for processing, segmenting, visualizing, registering, and analyzing medical images, as well as for image-guided treatments. Image segmentation, which focuses on identifying specific regions in the image such as tumors or lesions, is one of the most common challenges in medical image processing. In this research work we have utilized 3D Slicer to implement simulation techniques and automation for imaging diagnostics, computation, and prediction. This paper expands on the HD Brain Extraction module in 3D Slicer to autonomously segment brain MRI images using artificial intelligence. To optimize the brain extraction process, various adjustable parameters including segmentation techniques, threshold values, and smoothing factors are fine-tuned. The brain is then extracted from MRI images for further analysis and visualization
Multimodal Deep Learning for Disease Diagnosis and Risk Stratification: Integrating Genomic, Clinical, and Imaging Data
No full textPersonalized healthcare depends on the smart combination of heterogeneous biomedical information, including genomic sequences, clinical records, and medical imaging, so that it can be predictable with precision and interpretation. To accomplish this, the current study suggests a Hierarchy Attention Fusion based Multimodal Deep Learning (HAF-MDL) framework which improves the diagnostic accuracy and interpretability by intra- and inter-modality attention and Bayesian uncertainty measurement. In contrast to the conventional fusion methods, HAF-MDL learns the modality-relevant dynamically, avoiding uncertainty in heterogeneous patient data. To make the model clinical, it was trained and evaluated using a semi-synthetic dataset of 1,440 patient profiles in statistical agreement with real biomedical repositories TCGA (oncology), MIMIC-IV (clinical), and ADNI (neurology) to make the model clinically realistic. The Kolmogorov Smirnov (Ks) tests (p > 0.05) validation was also performed to ensure that the generated distributions were statistically consistent with real data in the world, which improved the reproducibility. The HAF-MDL framework proposed reached an accuracy of 94.8% and AUC of 0.964, which is higher than the unimodal and conventional fusion models. These results show that the suggested multimodal integration plan has great benefits in terms of the disease diagnosis and risk stratification and provides interpretability and reliability, generating a repeatable pathway to precision medicine
Evaluating Semi-Adaptive Signal Control Systems for Traffic Management: A Case Study of Key Intersections in Tirana
No full textTraffic is an integral part of urban life, playing a significant role in how we move and manage our daily activities. Every day, citizens of Tirana face severe traffic congestion, especially during peak hours, when the city\u27s roads are filled with vehicles, and long waiting times become a common phenomenon. This has caused serious problems for the city, including long delays, traffic congestion, and detrimental effects on quality of life and the environment. To address these challenges, it is crucial to explore and implement advanced traffic management systems. In this context, signal control methods, such as semi-adaptive and fully adaptive systems, offer an excellent opportunity to improve traffic flow and reduce delays and queues in the city. This paper assesses the effectiveness of a semi-adaptive traffic signal control system at major intersections in Tirana, utilising PTV Vissim simulation software. Results show a significant reduction in traffic delays and queue lengths, particularly during peak hours. The second part of this study will assess the performance improvements of this approach in terms of reduced congestion, enhanced traffic management, and overall network efficiency. The semi-adaptive system\u27s dynamic adjustment of signal phases based on real-time data leads to improved traffic flow and reduced congestion, providing a cost-effective solution for urban traffic management. The findings suggest that semi-adaptive systems can optimize traffic flow, particularly in growing cities that face congestion
Pumped-Storage Energy Systems for the Drin River Cascade: A Case Study
No full textThis article investigates the feasibility of implementing pumped-storage hydropower (PSH) systems within the existing hydropower plants (HPPs) of the Drin River cascade in Albania. Currently, five HPPs are operational along the Drin River, two of which are managed by private operators, while the remaining three are operated by the public utility “KESH sh.a”. Increasing climate variability necessitates adaptive operational measures to enhance the safety and resilience of these hydropower assets. In parallel, the rapid development of variable renewable energy sources, such as photovoltaic and wind power, has intensified interest in large-scale energy storage solutions. The study evaluates a case scenario involving the installation of a 200 MW pumped-storage system between two existing reservoirs in the Drin River cascade. Based on a hydrological year characterized by high rainfall, the results indicate that the proposed system could reduce downstream discharge by up to 30% while producing approximately 28 GWh of net electrical energy. Beyond energy generation, the assessment highlights broader system-level benefits, including improved flood risk management, enhanced climate resilience, and optimized hydropower operation. The findings suggest that pumped-storage integration could play a strategic role in supporting Albania’s energy transition and strengthening the operational flexibility of the Drin River hydropower system.
Energy Generation from Water Systems: A Technical and Cost-Benefit Analysis
Get PDFThe production of energy from the flow of water in drinking water supply pipes is an emerging field globally, and particularly novel in Kosovo. This method involves integrating water turbines with generators directly into water pipes, utilizing the water flow to rotate blades and drive the rotor, thereby producing electricity. Such systems not only generate renewable energy but also reduce excess pressure within the pipeline network, providing a dual benefit. Although similar technologies exist worldwide, their practical application in potable water systems remains limited, with experts yet to fully embrace their potential for reliable power generation. This paper explores the feasibility of implementing such a system in the Regional Water Company “PRISHTINA,” with the aim of using the generated electricity to power monitoring equipment in the water supply network. The proposed approach has the potential to enhance operational efficiency, generate additional revenue, and mitigate risks associated with high pipeline pressure. This paper provides novel insights into the technical, financial, and environmental benefits of harnessing energy from existing water distribution systems in underdeveloped regions
A Trimean and Asymmetry-Based Statistical Permutation Test for Group Comparisons
Get PDFStatistical multiple comparison tests are methods used to detect differences between several groups and to assess whether these differences are statistically significant. Typically, parametric tests like ANOVA (Analysis of Variance) or non-parametric tests like Kruskal-Wallis are employed for this purpose. In this study, we propose a new statistical permutation test based on the trimean and Bowley’s measure of asymmetry as an alternative to conventional multiple comparison tests. The proposed method is compared with ANOVA and Kruskal-Wallis tests in terms of reliability and statistical power. The analyses demonstrate that the proposed test yields statistically significant and effective results comparable to traditional methods. The findings reveal that the new test provides reliable outcomes especially for heterogeneous groups, skewed distributions, and small sample sizes. Overall, the proposed method can be considered a viable alternative in statistical analysis
Unsupervised Clustering of Multivariate Sports Activity Data Using K-Means: A Study on the Sport Data Multivariate Time Series Dataset
Get PDFThis work investigates the combination of unsupervised machine learning with blockchain- influenced data integrity aspects on multivariate time series (MTS) sports activity data. Using the SportData MTS dataset with complex physiological and movement parameters such as heart rate, speed, and altitude, we used K-Means clustering to uncover hidden patterns in the data and incorporated blockchain-influenced hash chains for traceability and integrity of data. Each of the datasets was standardized to ensure equal scaling, and three clusters were identified using silhouette score and elbow method evaluation. The result confirms K-Means to effectively cluster the data into tightly separated groups, with principal component analysis (PCA) plots confirming that there is substantial separation. Silhouette score analysis also confirmed the compactness and separability of groups. In addition, blockchain-inspired hashing was applied to each record to simulate tamper- evidence, providing a firm grounding for secure machine learning pipelines. The end-to-end solution not only reveals the inherent structure in sports activity data but also hints at maintaining data integrity to provide sound and transparent machine learning results, paving the way for future work in secure sports analytics, activity recognition, and anomaly detection
A Topic Modeling Analysis of Circular Economy and Big Data Research Using BERTopic and SciBERT
No full textThis study presents a hybrid topic modeling framework to map emerging themes in Circular Economy (CE)–Big Data literature. Using a corpus of 1,171 articles (2015–2025), three topic modeling techniques like BERTopic with SciBERT embeddings, Latent Dirichlet Allocation (LDA), and Top2Vec were applied and evaluated using coherence and diversity metrics. The transformer-based BERTopic–SciBERT model yielded 88 fine-grained topics with high coherence (mean Cᵥ= 0.47) and diversity (0.72), outperforming classical models in semantic quality and topic distinctness. Extracted topics were organized into five ontology-based domains: technical enablers, operational practices, policy/social, business models, and miscellaneous. Community detection in topic-similarity networks revealed distinct research clusters that moderately aligned with these ontology domains. Temporal analysis showed a structural shift after 2019, with increased focus on digitalization and data-driven sustainability. Policy-related themes remained limited, indicating gaps in governance research. Model robustness was validated through dimensionality sensitivity and embedding ablation, confirming stability and interpretability. A Sankey diagram was developed to visualize topic–domain–community linkages. The proposed framework provides a replicable method for semantic mapping in interdisciplinary sustainability research and supports strategic insight into evolving research directions in the CE–BD field