Hrčak - Portal of scientific journals of Croatia
Not a member yet
321498 research outputs found
Sort by
Who Did AI Leave Behind? Social Inequality Perceptions in the Use of AI Tools in Croatia
No full textGenerative artificial intelligence (AI) is increasingly embedded in everyday life, raising questions about how its use may reinforce or mitigate social inequalities. This study examines perceptions of affordability, self-assessed knowledge, practical accessibility, and usefulness of AI tools in Croatia, focusing on how gender, age, and frequency of AI use shape emerging digital divides. Drawing on survey data from a nationally representative sample, descriptive analyses, group comparisons, exploratory factor analysis, and multiple linear regressions were conducted to identify patterned inequalities across Lutz’s three sequential levels of digital inequality: access, skills, and outcomes. Factor analysis indicates that the inequality items do not form a single coherent scale, suggesting that AI-related inequality is multidimensional and that affordability, knowledge, practical accessibility, and usefulness represent distinct but related facets. Group comparisons and regression models reveal that frequency of AI use is the most consistent predictor across all facets: frequent users report higher affordability, greater perceived knowledge, lower reliance on assistance, and stronger perceptions of usefulness, while non-users cluster at the opposite end of each dimension. Age further differentiates respondents in perceived knowledge and practical accessibility, with younger cohorts feeling more competent and less dependent on help, whereas gender only marginally shapes confidence and loses significance once age and use frequency are controlled. Overall, the findings support and extend sequential models of digital inequality by demonstrating that, in the Croatian context, GenAI inequality is driven less by static sociodemographic attributes and more by practice-based divides between those who engage with AI tools and those who remain non-users
Relationship between destination competitiveness and behavioral intention: The case of agritourism destination
No full textPurpose - This study analyses the relationship between destination competitiveness
and behavioral intentions of agritourists in Prachinburi, Thailand. It investigates which
competitiveness dimensions most influence intentions to revisit and recommend. It also
explores the effects of sociodemographic as control variables.
Methodology/Design/Approach – Using models developed by Ritchie & Crouch (2010) and
Dwyer & Kim (2003), data from on-site agritourists was analysed using Factor Analysis and
Multiple Regression.
Findings –Results indicate that four competitiveness dimensions (accessibility, marketing
strategy, created resources, and endowed resources) positively influence revisit intentions. Five
dimensions (adding quality of service) influence recommendation intentions. Generational
cohorts and distance from destination affect competitiveness’s impact on revisit intentions,
while sociodemographic variables don’t influence recommendation intentions. The study
concludes that the role of supporting factors cannot be mitigated if a destination aims to
retain customers. Furthermore, destinations having less distinctive natural resources can still
be competitive by enhancing created resources and marketing strategy.
Originality of the research – This research contributes to literature by focusing on Asia Pacific/
developing countries, emphasizing the demand side, and exploring the agritourism context
Evaluation of microbiological safety and mycotoxin contamination of household produced meat products originating from Croatian indigenous pig breeds
No full textThis study aimed to assess the microbiological safety and mycotoxin contamination of traditional meat products made from Croatian indigenous pig breeds, produced in family households under uncontrolled conditions. Different factors, including natural, seasonal, and uncontrolled production conditions, pose challenges in maintaining consistent product microbiological and mycotoxin contamination. No bacteria from the genera Salmonella or Clostridium, or the species Staphylococcus aureus, were detected in any of the investigated samples. However, Listeria monocytogenes was found in a cured sausage sample from Black Slavonian pigs, rendering it unsafe for consumption. Additionally, Listeria innocua was identified in a dried sausage sample from Turopolje pigs. Yeast and mould contamination levels ranged as follows: cured sausages: 10² – 1.4 × 10⁴ cfu/g; whole ham: 1.4 × 10² – 2.7 × 10⁴ cfu/g; bacon: 2.6 × 10² – 3.5 × 10³ cfu/g. The dominant mould genus colonising the dry meat products was Penicillium (30 isolates), followed by Aspergillus (20 isolates), with Cladosporium and Mucor species present in lower numbers. The most frequently isolated Penicillium species were P. brevicompactum, P. commune, and P. solitum (85.7%), while the most common Aspergillus species were A. proliferans and A. tubingensis (57.1%). Regarding product type, bacon and ham met the respective safety standards, but sausages were contaminated with L. monocytogenes and L. innocua, making them unsafe for consumption. All products were safe in terms of mycotoxin contamination
Hybrid Renewable Energy Systems for Public Buildings: Optimization and Sustainability
No full textThis study develops an optimization model for hybrid renewable energy systems (HRES) to improve energy efficiency and sustainability in public buildings. A case study at the Faculty of Mechanical Engineering in Mostar integrates a water-to-water heat pump, photovoltaic panels, solar collectors, and a small wind turbine. A two-step optimization process was used: the Simplex method optimized the thermotechnical system, minimizing the Net Present Cost (NPC), while Homer Pro determined the optimal electricity generation configuration. The VIKOR method ranked six scenarios based on energy, environmental, and economic criteria. Results show that Scenario S5, featuring a 120 kW water-to-water heat pump and solar collectors covering 60% of domestic hot water demand, achieved the lowest primary energy consumption (58,438.6 kWh), lowest CO2 emissions (19,284.7 kg/year), and the most favourable Levelized Cost of Heating/Cooling (LCoH/C). Scenario S2, with pellet and air-to-water heat pumps, exhibited the highest autonomy (100%) but at a higher cost. The study confirms that HRES, particularly heat pump-based solutions, can cut emissions and reduce reliance on fossil fuels while maintaining economic feasibility. These findings support the transition towards prosumer-based energy models, aligning with Bosnia and Herzegovina’s decarbonization goals
Optimized Deep Learning Model Using Capsule Networks and GRUs for Predictive Analytics in Smart Cities
No full textSmart cities rely on accurate predictive models to enhance sustainability, optimize resource utilization, and improve the overall quality of urban life. Forecasting key parameters such as air quality, water quality, and waste generation is crucial for efficient urban planning and environmental management. This study proposes a novel deep learning-based prediction model that integrates Capsule Networks and Gated Recurrent Units (GRUs) for multidimensional forecasting in smart cities. The architecture begins with a one-dimensional convolutional layer (Conv1D) to process raw input data, followed by a Capsule layer for efficient spatial feature extraction. These spatial features are subsequently fed into stacked GRU layers to model temporal dependencies and improve predictive accuracy. To further enhance the model’s performance, the Osprey Optimization Algorithm (OOA) is employed for hyperparameter tuning. The proposed hybrid model demonstrates superior forecasting capabilities when evaluated using standard metrics. It achieves a Mean Squared Error (MSE) of 43.8751, Root Mean Square Error (RMSE) of 4.4083, R² Score of 0.6845, and Mean Absolute Error (MAE) of 2.1619. These results highlight the effectiveness of combining Capsule Networks and GRUs, along with OOA optimization, for robust and accurate predictive modeling in smart city applications. This model serves as a scalable and efficient tool for urban administrators and policy-makers to enable data-driven decisions and proactive management of city resources
Integrating Sustainability at the Project Lifecycle's Strategic Entry
No full textIn response to increasing demands for sustainable development, organizations are progressively integrating sustainability principles into project management practices. This study focuses on the Initiation Phase of the project lifecycle, a critical juncture for embedding environmental, social, and governance (ESG) considerations. Drawing on data from 754 respondents across diverse industries, the study examines current practices, competency gaps, and organisational readiness for sustainable project delivery. Quantitative analysis of survey responses reveals significant discrepancies between the perceived importance of sustainability competencies and their implementation during project initiation. Key gaps include defining sustainability-focused objectives, assessing ESG impacts, and embedding governance mechanisms. Although some organizations demonstrate maturity in strategic ESG integration, findings indicate that many still rely on ad hoc or reactive approaches, often lacking formal roles or clear structures for sustainability. The study underscores the need for dedicated sustainability roles, targeted training, and stronger alignment of governance frameworks with ESG goals. By highlighting the systemic nature of competency gaps and the critical role of organizational support, this research offers practical insights for advancing sustainable project management from the very outset of project planning. The findings demonstrate that ESG-related competency gaps at the project initiation phase are largely systemic rather than incidental, stemming from misalignment between strategic intentions and organisational readiness
Investigation of the Influence of Machining Parameters on Surface Roughness in Turning Operations and Machine Learning Application
No full textThe performance of turning operations gradually depends on the machining parameters, and the most important parameter is the surface roughness quality. In this study, the effect of cutting speed and feed rate on the surface roughness in aluminium alloy 6082 (AA6082) machining, which is widely used in the automotive and aerospace sectors, was investigated experimentally and by machine learning prediction. In the experiments, three different cutting speeds (240, 300, and 360 m/min), three different feed rates (0.05, 0.1, and 0.15 mm/rev), and a constant depth of cut (0.5 mm) were used as machining parameters. In addition to machining parameters, the temperature, cutting forces, revolution, current, voltage, and power were measured. The workpiece was machined using uncoated cemented carbide cutting tools. Experimental results showed that the surface roughness increased with increasing feed rate and decreased with increasing cutting speed. The complete dataset was created from experiments by selecting measurements and machining parameters as inputs and surface roughness as output. Various machine learning models were implemented on this dataset, and different metric scores were used to select the best prediction performance of the machine learning models. Gradient Boosting (GB) exhibited superior prediction performance compared to the other tested algorithms, with an R2 score of 0.98560. The GB model emerged as the most precise and accurate, characterized by the highest R2 score, the lowest root mean squared error (0.12095), the lowest mean absolute error (0.09804), and the lowest mean squared error (0.01463) scores, respectively
Estimation of Rock Brittleness from Point Load Strength Index Data Using Machine Learning Methods
No full textBrittleness is a vital mechanical property that characterizes a rock's tendency to fracture under applied stress without significant deformation, which is particularly significant in mining, tunnelling, and other geotechnical engineering applications. The accurate prediction of rock brittleness is essential for optimizing excavation strategies, ensuring operational safety, and improving the cost-efficiency of resource extraction processes. However, conventional brittleness assessment techniques-such as those based on uniaxial compressive strength (UCS) and tensile strength-can be labour-intensive, time-consuming, and expensive. This study introduces a predictive framework based on machine learning algorithms using Point Load Strength Index (PLI) values as the sole input variable. A comprehensive dataset comprising sedimentary, igneous, and metamorphic rocks was compiled from both literature sources and laboratory experiments. Multiple regression models were applied and compared, including traditional linear methods and advanced ensemble learners. Among these, the Gradient Boosting Regressor delivered the highest predictive accuracy, achieving an (R²) value of 0.96 for metamorphic rocks. The results demonstrate that even a single indirect measurement like PLI can serve as an effective predictor of rock brittleness when coupled with robust machine learning techniques. The findings highlight the potential of integrating AI-based models into rock mechanics workflows to streamline brittleness estimation and support sustainable mining practices
The myth of normality testing in biomedical research
No full textTesting data for normality before applying parametric statistics has become a routine procedure in biomedical research. This commentary argues that such tests provide little inferential value, may mislead analytical choices, and reflect outdated thinking from a pre-computational era. Parametric procedures are robust to modest departures from normality, and the Central Limit Theorem makes most normality checks unnecessary. According to this Theorem, the sampling distribution of the mean approaches a normal shape as sample size increases, regardless of the original distribution of the data. Outlier panic and indiscriminate data ranking further undermine the meaning of measurement and prediction. The obsession with distributional purity has replaced the logic of inference with statistical rigor. It is time to abandon this ritual and refocus on design, representativeness, and modeling – the true pillars of inference
Application of multi-algorithm approach for lung cancer prediction
No full textLung cancer is one of the leading causes of cancer-related mortality worldwide, with most cases diagnosed at an advanced stage. Accurate and cost-effective early detection remains a major challenge due to the heterogeneity of imaging and histopathological features. Therefore, this study aimed to develop diagnostic software for lung cancer prediction using a multi- algorithm method. Patient data, including 16 clinical and lifestyle variables, were processed and analyzed with five machine learning algorithms, namely Neural Network (NN), Support Vector Machine (SVM), k-Nearest Neighbors (k-NN), Random Forest (RF), and Naïve Bayes (NB). Model performance was evaluated based on accuracy, precision, recall, and F1-score. The results showed that RF, NB, SVM, and NN achieved perfect predictive performance (100% across all metrics), while k-NN obtained slightly lower but still high performance (99%). These findings signified that multi-algorithm predictive modeling could provide robust diagnostic support for lung cancer detection. The proposed software offered potential as an accessible, low-cost decision-support tool to assist clinicians in early diagnosis and improve patient outcomes