TULTECH Journals
Not a member yet
235 research outputs found
Sort by
Data-Driven Predictive Modelling of Employee Absenteeism Using Workflow Automation Platforms
No full textEmployee absence is a critical factor affecting organizational productivity and employee well-being. This study presents a data-driven predictive framework for employee absenteeism using a newly collected enterprise dataset comprising 8,336 employees. Absenteeism is formulated as a binary classification task, distinguishing employees with more than 80 hours of annual absence from those with lower absence levels, based on demographic and occupational characteristics. The proposed approach applies gradient-boosted decision tree models, including LightGBM, XGBoost, and CatBoost, evaluated through a stratified train–test split at the employee level to approximate temporal separation between training and prediction. Feature engineering procedures are detailed, including categorical encoding and the construction of a commuting-related indicator. All models demonstrate strong predictive performance, achieving accuracy between 85% and 87%, precision ranging from 78% to 80%, recall between 76% and 79%, and AUC–ROC values of 0.92–0.93. Model interpretability is addressed using SHAP-based feature attribution, identifying age, gender, and occupational role and location as key predictors of absenteeism risk. Furthermore, a practical system architecture is outlined, integrating the predictive models within an automated workflow using the n8n orchestration platform for deployment in human resource information systems. This enables proactive identification of high-risk absenteeism cases and supports early intervention strategies with minimal human oversight. The study contributes by addressing data leakage concerns, improving feature transparency, and demonstrating a deployable and interpretable predictive system. Future research directions include multi-organizational validation, temporal modelling using sequential data, and evaluation of system-level effectiveness in real-world HR settings
Seismic Performance and Retrofit-Based Guidelines for DS-4 Unreinforced Masonry Buildings Evidence from Eight Real Case Studies
No full textUnreinforced masonry (URM) buildings constitute a substantial share of the existing building stock in many seismic regions and have repeatedly demonstrated poor seismic performance during recent earthquakes. After strong ground motions, a significant portion of these structures are classified as severely damaged (DS-4), a condition that typically requires detailed structural assessment and retrofit in order to restore acceptable safety levels. Although the seismic behaviour of URM buildings has been widely investigated, practical performance ranges derived from real post-earthquake case studies remain limited, particularly for buildings already affected by severe damage. This study presents a seismic performance assessment of eight real URM buildings classified as DS-4 following the 26 November 2019 Albania earthquake. All buildings were subjected to detailed on-site inspections, material characterization through laboratory testing, numerical modelling, and nonlinear static (pushover) analysis. The seismic response was evaluated both in the damaged (existing) condition and after the implementation of retrofit interventions designed in accordance with Eurocode 8 – Part 3. A macro-element modelling approach was adopted to ensure consistency in the evaluation of global capacity, drift limits, and ductility across the dataset. The results provide quantitative insight into the typical seismic capacity of severely damaged URM buildings and demonstrate the level of performance improvement achievable through commonly adopted retrofit strategies. Based on the synthesis of the eight case studies, realistic seismic performance ranges for DS-4 URM buildings are identified, together with target values for retrofitted configurations. The findings are intended to support engineers involved in post-earthquake assessment and retrofit design by providing reference benchmarks grounded in real building behaviour
Maritime Accidents in Albania: An Empirical Survey-Based Analysis
No full textMaritime accidents have long been a subject of concern due to their severe human, environmental, and economic consequences, including loss of life, environmental pollution, and disruption of global supply chains. These impacts underline the need for a deeper understanding of the factors contributing to maritime accidents, particularly the complex role of human error shaped by cognitive limitations, organizational practices, and environmental conditions. This study develops an empirical model linking Human Resource Management (HRM) practices to maritime accident frequency using count-data regression techniques. Empirical evidence is drawn from a structured questionnaire administered to 68 maritime professionals employed in maritime organizations and companies, including crew members on vessels calling at national ports. The analysis examines organizational characteristics, accident and incident reporting practices, accident frequency over the previous five years, and perceptions of safety and operational management. Results show that 82.5% of respondents work in organizations that systematically collect evidence following accidents or incidents. On average, respondents reported 7.33 accidents over the five-year period. The regression results indicate that stronger HRM practices are associated with a significant reduction in expected accident frequency, even after controlling for experience and training. These findings highlight the importance of organizational learning, management quality, and system-level prevention strategies in enhancing maritime safety
Data-Driven Regression Modelling of Insolvency Outcomes: Judicial Efficiency, Foreign Participation, and Recovery Trends
No full textThis paper provides an in-depth analysis of Albania’s debt relief system over the period 1995–2020. The dataset comprises annual bankruptcy court caseloads, foreign-creditor participation rates, and asset-reclamation statistics, allowing an examination of long-term trends amid substantial fluctuations. Using segmented time-series regression, we identify significant structural turning points associated with two major legislative reforms in 2002 and 2016. Both the level and slope of annual bankruptcy filings increased markedly following these legal interventions. A fractional logistic model indicates that foreign-creditor involvement, consistently between 13% and 15%, increased notably after 2016, reflecting improved functionality of the international recognition system for cross-border operations. Asset-recovery rates, averaging 59%, were analysed using regression and Generalized Additive Models (GAMs), showing that recovery efficiency declines under heavier judicial caseloads but improves in years with greater foreign participation. Comparative analysis with Romania, Bulgaria, and Serbia demonstrates that Albania is approaching regional norms, though gaps remain between recovery performance and institutional capacity. Overall, the results highlight those judicial reforms, the use of statistical tools in administrative decision-making, and the combination of legal modernization, courtroom efficiency, and international integration are critical determinants of effective bankruptcy systems
Quantum Machine Learning Algorithms for Optimizing Complex Data Classification Tasks
No full textQuantum Machine Learning (QML) has emerged as a paradigm that combines the computational advantages of quantum computing with the predictive capabilities of machine learning to address complex data classification problems. As data dimensionality increases rapidly and classical learning algorithms face scalability constraints, QML leverages quantum parallelism, entanglement, and high-dimensional Hilbert space representations to enhance learning performance. This paper reviews and analyses advanced QML algorithms, including Quantum Support Vector Machines (QSVM), Variational Quantum Circuits (VQC), Quantum Neural Networks (QNN), and Quantum Kernel Estimation, for optimizing both binary and multi-class classification tasks under high-complexity conditions. The proposed QML framework is evaluated on benchmark datasets like MNIST, the Breast Cancer Wisconsin (BCW) dataset, and synthetic nonlinear datasets, and is compared against classical machine learning baselines, including Support Vector Machines (SVM), Random Forests (RF), and Deep Neural Networks (DNN). The results demonstrate notable improvements in classification accuracy (up to 96.8%), decision margins, and computational efficiency in quantum-suitable data regimes, while also highlighting current limitations related to noise, circuit depth, and hardware constraints. Overall, the study presents a unified QML framework, theoretical formulations, and experimental evaluations that illustrate the potential of quantum algorithms for next-generation classification tasks
Stacking Ensemble Deep Neural Networks with Regressor Chains for Building Energy Performance Prediction
No full textThe energy performance of buildings (EPB) is a critical factor in reducing global energy consumption, mitigating greenhouse gas emissions, and achieving sustainability goals. Predictive modelling of EPB constitutes a complex, non-linear multi-target learning problem, where multiple continuous outputs must be estimated simultaneously from a common set of input variables. Multi-Target Regression (MTR) presents significant challenges due to complex output dependencies, high output dimensionality, imbalanced and noisy targets, and distributional shifts, which collectively degrade predictive performance. To address these challenges, this study proposes a novel ensemble regressor-chain framework integrated with a stacking ensemble deep neural network architecture for MTR modelling. The proposed approach is evaluated using five benchmark multi-target regression datasets related to building energy performance. Experimental results demonstrate that the proposed model consistently outperforms classical regression methods (linear regression, generalized linear models, and CART) as well as recent state-of-the-art approaches, including regression forests and sparse regression techniques. Performance gains of up to 12% reduction in RMSE and a 9% improvement in R² are achieved. Robustness is further validated through statistical testing using the Friedman test with Finner’s post-hoc correction, supported by visual analyses such as scatter plots and error distributions. Overall, the results indicate that ensemble deep learning architectures combined with regressor chains provide a more effective and scalable solution for multi-target EPB prediction than traditional regression models, offering practical value for real-world energy efficiency assessment and sustainability-oriented decision making
A Data-Driven Survival Analysis of Prognostic Determinants in Patients with Alcohol-Related Liver Disease: A Prospective Study
No full textAlcohol-related liver disease (ALD) is a leading cause of liver-related mortality in Europe, yet prospective survival data from Southeast Europe remain limited. Prognostic assessment has traditionally focused on biological disease severity, while behavioral factors particularly sustained alcohol abstinence is less consistently incorporated. It has been conducted a prospective observational cohort study of 200 adults with confirmed ALD treated at a national tertiary referral center in Albania and followed for 12 months. Sustained alcohol abstinence (≥6 months) was modelled dynamically as a time-varying exposure within an integrated biological–behavioral prognostic framework. Overall survival was evaluated using Kaplan–Meier analysis and Cox proportional hazards models, with liver transplantation treated as a censoring event; competing-risk models were applied to account for transplantation as a competing outcome. During follow-up, 44 patients (22%) died. Non-survivors had significantly higher Model for End-Stage Liver Disease (MELD) scores (21.0 ± 7.1 vs. 15.0 ± 6.2, p < 0.001) and a higher prevalence of ascites (77% vs. 46%, p = 0.002) and hepatic encephalopathy (52% vs. 19%, p < 0.001). Sustained abstinence was less frequent among non-survivors (20% vs. 46%, p = 0.013) and was associated with improved survival (log-rank p = 0.013). In multivariable Cox and competing-risk analyses, MELD, ascites, and hepatic encephalopathy independently predicted mortality, whereas time-varying abstinence demonstrated an independent protective effect. The combined biological–behavioral model showed good discrimination and calibration (optimism-corrected Harrell’s C-index 0.78–0.82; 12-month AUC ≈ 0.80). In this underrepresented Southeast European cohort, established severity markers remained dominant predictors of short-term mortality, while the dynamic incorporation of abstinence provided incremental prognostic value, supporting improved risk stratification and pragmatic ALD management in resource-limited settings
Quantifying Teachers’ Knowledge and Attitudes Toward Personal Data Protection Using Regression Models
No full textThe use of digital platforms and mobile applications in schools has increased substantially, leading to a corresponding rise in the volume of personal data processed for educational purposes and highlighting the need to strengthen privacy awareness and cybersecurity practices among staff. This study examines teachers’ awareness, attitudes, and experiences regarding personal data protection in educational institutions in Kosovo using a structured questionnaire (N = 60). Instrument reliability was confirmed using Cronbach’s alpha (α = 0.715), while the suitability of multivariate analysis was supported by sampling adequacy and correlation structure (KMO = 0.679; Bartlett’s test p = 0.001). Two regression models were employed to assess the impact of educational level on (i) teachers’ knowledge of personal data protection and (ii) the perceived importance of data protection. Results indicate that both outcomes increase linearly and significantly with educational level, suggesting that higher educational attainment is associated with greater awareness and stronger valuation of data protection. Additionally, a binary incident model and association tests were used to examine gender differences in reported data-related incidents, revealing higher odds of incident reporting among male participants and significant disparities across attitude–behaviour indicators. Overall, the findings underscore the importance of systematic training and clear institutional policies to support secure data-handling practices and safeguard student privacy in digital learning environments
Multi-Task Deep Learning Framework for Segmentation and Severity Estimation of Leaf Diseases in Multi-Crop Environments
No full textCrop diseases pose a major threat to global food security, creating a pressing need for effective and accurate diagnostic mechanisms that can be applied across diverse agricultural settings. This paper proposes a Multi-Task Deep Learning Framework (MTDLF) for the simultaneous segmentation of diseased regions and estimation of disease severity in crop leaves. The framework employs a shared ResNet-50 encoder with two task-specific decoders: a U-Net-based segmentation branch and a regression-based severity prediction head, trained using a composite loss formulation. In addition to the dual-task architecture, two consistency-driven mechanisms are introduced. A Severity-Constrained Segmentation Refinement (SCSR) module aligns predicted lesion-area proportions with estimated severity values, while a Lesion-Area Distribution Matching (LADM) loss enforces distributional consistency between segmentation outputs and severity-based lesion expectations. The model is trained and evaluated on publicly available, severity-annotated datasets of rice, maize, tomato, grape, and cotton leaves. Experimental results demonstrate that the proposed framework achieves a mean Intersection over Union (IoU) of 85.7%, a Dice coefficient of 88.3%, a Mean Absolute Error (MAE) of 7.5, and an of 0.92, outperforming conventional single-task methods and recent multi-task baselines. Furthermore, the model attains real-time inference performance of approximately 25 ms per image, making it suitable for edge-level deployment. The proposed MTDLF provides a unified and efficient approach to multi-crop disease monitoring, offering a practical pathway toward reliable, data-driven precision agriculture
Charting the Digital Frontier: A Comprehensive Bibliometric Analysis of E-Agriculture Research
No full textThis study presents a statistically validated bibliometric analysis of e-agriculture research published between 2020 and 2025, based on 1,363 peer-reviewed articles indexed in Scopus and Web of Science, and selected according to the PRISMA 2020 guidelines. Bibliometric mapping is combined with inferential statistical analysis and network validation to examine publication dynamics, thematic evolution, citation impact, and global collaboration patterns. Results show rapid growth in research output up to 2023, followed by a contraction in 2024. Core research themes include smart farming, Internet of Things (IoT), artificial intelligence particularly deep learning and precision agriculture. While China, India, and Brazil lead in publication volume, the United States, the Netherlands, and Germany exhibit higher citation impact, indicating a divergence between productivity and influence. Inferential testing confirms these patterns: one-way ANOVA reveals significant temporal variation in citation impact (F(5,1357)=48.5, p<2×10⁻¹⁶), and network modularity analysis (Q=0.519) demonstrates a robust thematic structure. Poisson regression further shows that publication year and thematic focus jointly shape citation performance. To extend beyond descriptive bibliometrics, the study integrates an altmetric perspective, drawing on Twitter sentiment and topic analysis to capture societal engagement with digital agriculture research. Overall, the study advances bibliometric analysis in e-agriculture by combining statistical validation, network robustness assessment, and signals of societal impact