Various frameworks for IT governance have been proposed by academicians, the business world, and professional associations. However, no common framework has been proposed for any individual components of IT governance in the literature. The paper developed a comprehensive framework for IT governance domains, one of which is the components of the abovementioned framework. In the relevant literature, especially those who propose a framework have been emphasized, and not only the academic world but also the business world have been considered. This study was built on two interacting foundations: review and design. The domain names, each representing an entity, were obtained from a selected literature review on IT governance domains and contexts and used as data or input for the classification and abstraction study. A comparative and analytical classification and abstraction study was conducted on this dataset. As a result of the reasoning and comparison process, four types of domains have been established. The proposed abstract framework was implemented on the reviewed literature. This flexible and comprehensive framework provides a contextual basis for both researchers and practitioners and fills the gap in the framework for IT governance domains. Furthermore, this framework will assist in utilizing the power of IT in achieving enterprise objectives. The proposed component-based
Escalating global disasters demand a strategic transition from static mapping to dynamic intelligence in crisis management. This bibliometric analysis maps the evolutionary trajectory of GIS in disaster research, using a dataset of 3,061 publications from the Web of Science (1993-2025). The findings reveal a profound epistemic transition: the field has evolved from an initial phase of "Reactive Baselining" (1993-2010) to the current era of "Disaster Intelligence" (2019-2025), driven by the integration of AI, deep learning, and real-time spatiotemporal analytics. However, the analysis also uncovers a polarized knowledge landscape where innovation is heavily concentrated in China, the United States, and Western Europe, revealing a significant "digital divide" that restricts the transferability of advanced geospatial tools to under-resourced regions in the Global South. Furthermore, the operationalization oftechnologiessuch as digital twins is hindered by interoperability and ethical data governance challenges. Policy implications are critical; to bridge the gap between algorithmic potential and on-the-ground reality, stakeholders must prioritize equitable capacity building, standardize cross-border data protocols, and establish rigorous privacy frameworks. This study provides a strategic roadmap for leveraging next-generation geospatial solutions to foster adaptive, resilient, and ethically governed global disaster management.
Customer feedback is a critical asset for navigating competitive markets. This study aims to analyze online customer complaints in the road passenger transport sector to identify service quality dimensions and pinpoint recurring industry issues, thereby offering strategic insights for enhancing customer satisfaction. The research used a comprehensive dataset of 7,719 complaints regarding the three most preferred transport companies in Turkey, retrieved from www.sikayetvar.com. Data analysis was conducted using Python-based text mining techniques, specifically latent Dirichlet allocation (LDA) for unsupervised topic modeling and classification. The findings revealed that complaints primarily focused on physical characteristics, trust-related issues, and the physical conditions of vehicles. Comparative analysis showed that while Company K and Company P received significant complaints in the "Tangibility" dimension (30.46% and 24.4%, respectively), Company M faced its highest complaint rate in the "Reliability" dimension with 58.93%. These results suggest that service quality gaps vary by company, requiring tailored improvements in physical assets or service reliability to boost overall passenger satisfaction.
Audiovisual techniques such as highly saturated colors, rapid editing transitions, and elevated audio energy are widely employed to maximize viewer engagement in children’s media content. These productions, which fall outside conventional mainstream narrative techniques, are referred to as hyperstimuli and may encourage children to remain in front of screens for extended periods, potentially leading to various health-related problems. This study examined 90 videos with the highest view counts from 30 popular children’s YouTube channels and aimed to analyze their structural characteristics using a quantitative, comparative research design. For this purpose, a custom software tool was developed using the Python programming language, incorporating the OpenCV, NumPy, Librosa, FFmpeg, csv, and os libraries. The software sampled one frame per second and calculated the mean saturation (SAT), mean brightness/value (VAL), colorfulness (COL), visual complexity (ENT), cuts per minute (CPM), audio energy (AE), and audio spectral centroid (AC). These variables were transformed into z-scores, standardized, and combined by computing an equally weighted average to construct a comprehensive hyperstimulation index (HI) using IBM SPSS Statistics 26.0. This approach enabled the analysis of hyperstimulation levels in the sampled content based on quantitative parameters. The findings indicate that rapid editing transitions (CPM) and high audio energy (AE) constitute the strongest components of the hyperstimulation index, and view counts increase markedly as hyperstimulation levels rise. Overall, the results indicate that popular children’s YouTube channels favor narrative structures characterized by high stimulus intensity in alignment with the logic of the attention economy and that higher viewing levels are associated with this preference.
Forest fires are among the most destructive natural disasters, causing substantial ecological, economic, and human losses. Accurate assessment of fire severity is crucial for preparedness, rapid response, and efficient resource management. This study evaluates three supervised machine learning (ML) algorithms -Linear Discriminant Analysis (LDA), Kernel Naive Bayes (KNB), and Fine Gaussian Support Vector Machine (Fine Gaussian SVM)-to classify forest fire severity using a real-world dataset from T & uuml;rkiye. The dataset includes over 15,000 fire incidents (2010-2024) and 36 initial features. To improve predictive performance and reduce dimensionality, feature selection was performed using the Chi-square test. Fire severity was reclassified into three levels (low, moderate, high) based on burned area (hectares). Models were trained and validated with 10-fold cross-validation. KNB achieved the highest accuracy (82%), followed by Fine Gaussian SVM (79%) and LDA (65%). The advantage of KNB likely stems from its ability to capture nonlinear class boundaries and probabilistic structures typical of complex environmental data. Overall, the results suggest that nonlinear, kernel-based classifiers outperform linear methods for forest fire severity classification. The proposed national-scale, interpretable framework can support policymakers and disaster-management authorities in developing intelligent early warning systems and optimizing suppression resource allocation in high-risk areas.
Venture capital (VC) is increasingly adopting artificial intelligence (AI) strengthen data-driven decision-making. This article investigates how AI is reshaping VC investment practices, drawing on a survey of 15 professionals—including emerging fund managers, venture analysts, and asset managers—from Turkey, Lithuania, and neighboring emerging ecosystems. We present the survey methodology, review academic and industry perspectives, and highlight key findings on AI adoption, perceived benefits, barriers, and future outlook.The survey included questions on current AI use, benefits, barriers, and expectations. The results show that most respondents already use AI tools, primarily to enhance deal sourcing and due diligence efficiency. Participants identified faster data processing, improved decision quality, and reduced human bias as major benefits, echoing prior research, while citing data availability, explainability, and cultural resistance as critical barriers. In the future, respondents expect AI to become integral to VC, a competitive advantage for those who embrace it, while emphasizing that human judgment and oversight will remain essential. Overall, AI emerges as a transformative force in venture investing, augmenting human expertise and offering the potential to significantly improve efficiency and generate deeper insights across decision-making workflows, provided that the challenges of data quality, transparency, and human–AI collaboration are effectively addressed.This study contributes a novel, practice-grounded perspective by presenting original survey evidence from emerging venture ecosystems (Türkiye and Lithuania), a context largely absent from prior academic work that has focused on large, mature markets. The paper thus extends the evidence base on how investors actually employ AI across the venture lifecycle.
Employing the systematic review methodology, this study aims to address how digital health literacy and e-health literacy influence people’s quality of life. This systematic review includes studies that examined the effect of e-health literacy and digital health literacy on quality of life. The results of 12 research papers meeting the standards established within the parameters of this study were divided into two groups: e-health literacy and digital health literacy. The effect of e-health literacy on quality of life was assessed in five of the thirteen studies, whereas that of digital health literacy on quality of life was assessed in eight of them. The study sample consisted of people with cancer, people with long-term conditions like asthma, people who were following their postoperative treatment regimens, people affected by the COVID-19 epidemic, the elderly, women, children, and students. Only one study reported that e-health literacy had a negative influence on quality of life, whereas 11 studies emphasized that digital health and e-health literacy had a positive influence. Policies are required to facilitate the use and access to digital health technology by various societal groups, including patients, the elderly, women with unequal social status, children, and learners.
Accurate segmentation of lung and infection regions in CT scans plays a vital role in the early diagnosis and clinical management of COVID-19. This study proposes a robust and efficient DL approach utilizing a U-Net architecture equipped with ResNet and SEResNet encoder backbones, combined with optimized preprocessing strategies. The integration of flipping and sliding window techniques, which substantially increased the data volume and enhanced feature extraction at multiple scales, is a key innovation of this work. Unlike many prior studies that focused solely on infection segmentation, this study addresses both lung and infection region segmentation on the publicly available COVID-19 CT dataset. Among the various backbones tested, SEResNet18 was selected as the optimal choice due to its high accuracy and computational efficiency.The proposed model achieved outstanding segmentation performance, with Dice score, F1-score, and IoU reaching 0.9897, 0.9900, and 0.9796 for lung segmentation and 0.9339, 0.9357, and 0.8770 for infection segmentation, respectively. These results not only surpass those of previous studies using the same dataset but also highlight the significance of TP in improving model generalizability. This study contributes to the field by demonstrating that carefully designed data preparation pipelines can be as impactful as architectural innovations, paving the way for high-performance, resource-efficient segmentation systems applicable to broader medical imaging tasks.
Datasets play a crucial role in advancing NLP tasks by providing resources for training and evaluating models. These datasets serve as the foundation for developing models that can understand and generate human language. However, little research has been performed on creating datasets in low-resource languages, such as Turkish, compared to English. This research focuses on paraphrase datasets, which is a limited area of study in NLP. To the best of our knowledge, nine Turkish paraphrase datasets are available. Accordingly, we provide a comprehensive review of Turkish paraphrase datasets by examining their scope, design, and linguistic properties. The pros and cons of the datasets are discussed, including their size, variety, and domain coverage limitations. Suggestions are presented for a more diverse, large-scale, and expandable Turkish paraphrase corpora that could better represent the language’s linguistic richness.
Low back pain (LBP) is one of the leading causes of disability worldwide, posing substantial clinical and socioeconomic challenges. Machine learning (ML) has introduced new opportunities for improving early diagnosis, risk stratification, and treatment guidance in LBP management. This systematic review aims to synthesize current evidence on the application of ML techniques in the diagnosis and clinical management of LBP and to identify methodological patterns and performance trends.Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) framework, 27 peer-reviewed studies published between 2017 and 2023 were systematically identified and evaluated according to predefined inclusion and exclusion criteria.The findings indicate that Convolutional Neural Networks (CNNs) achieve high accuracy rates (98%–99%) in large-scale and imaging-based datasets, such as magnetic resonance imaging (MRI) and EHR data. SVMs demonstrate strong predictive performance in real-time sensor signal analysis, while ensemble methods, including Random Forest and XGBoost, yield robust results in structured clinical and biomarker datasets. Nevertheless, substantial methodological barriers remain. These include data heterogeneity and noise in PROM–based datasets, performance decline during external validation in real-world clinical settings, and limited interpretability due to the black-box nature of many ML models.This review highlights the need for multimodal data fusion strategies and the integration of Explainable Artificial Intelligence (XAI) frameworks to enhance model transparency, generalizability, and clinical trust. Advancing these dimensions is critical for translating ML–based LBP models from experimental settings into routine clinical decision support systems and for shifting healthcare delivery from reactive management to proactive, data-driven care ecosystems.
Monitoring soil organic matter (SOM) is a cornerstone of sustainable agriculture and food authenticity verification, yet traditional laboratory-based assessments are too resource-intensive for regional-scale deployment. This study presents a rigorous comparative evaluation of two sensing modalities: a terrestrial IoT sensor network and polarimetric synthetic aperture radar (PolSAR). To address the lack of laboratory-certified labels, we engineered the Soil FertilityIndex (SFI)and Component-based Soil QualityIndex (CSQI) as continuous biophysical proxies. Our methodology employs a dual-phase machine learning strategy: first, unsupervised latent discovery was used to identify dominant geophysical regimes within the radar backscatter. Second, an ensemble of supervised regression algorithms, including XGBoost and LightGBM, was deployed using a 10-fold cross-validation protocol to estimate soil integrity across the study site. The Geophysical Handshake novelty, which mathematically harmonizes ground-level dielectric measurements with satellite-based polarimetric scattering signatures, is central to our framework. To ensure the resilience of our findings, we implemented a strict identity-feature audit to eliminate data leakage. Results reveal a significant performance disparity: while the IoT track provided stable local insights, the PolSAR-based models demonstrated superior predictive power, achieving a Blind-Test R-2 of 0.9908 and an RMSE of 0.33, with a negligible cross-validation standard deviation of 0.0003. This study proves that the integration of continuous proxy engineering and supervised radar modeling provides the synoptic, vegetation-penetrative X-ray required for resilient and scalable SOM estimation across non-instrumented landscapes.
The selection of the right location is vital for the survival and growth of new business ventures, yet entrepreneurs often lack reliable and localized market insights to guide such decisions. To address this challenge, this study introduces Powllster, a gamified mobile crowdsourcing platform designed to help entrepreneurs and businesses evaluate potential locations and market opportunities. By combining crowdsourcing and gamification principles, the platform encourages users to provide location-based evaluations of businesses and neighbourhoods in an engaging manner. Usability and effectiveness were tested using the System Usability Scale (SUS) and Net Promoter Score (NPS). The results revealed an SUS score of 73.27 and an NPS score of 18.51, reflecting positive user sentiment and a moderate willingness to recommend. The participants highlighted ease of use, seamless functional integration, and confidence in interaction, whereas negative perceptions were minimal. These findings demonstrate that Powllster is intuitive, engaging, and valuable for collecting and accessing market intelligence. In practice, it lowers barriers to market research and provides entrepreneurs with actionable, crowdsourced insights for data-driven decisions. Unlike traditional decision-support systems, Powllster integrates participatory data collection with user engagement, offering a novel and scalable tool for guiding new business ventures.
While the dark web is designed to protect user privacy, it is also susceptible to misuse by malicious actors. Therefore, the effective classification of dark web (Tor and NonTor) traffic is of significant importance in cybersecurity. Findings from the literature reveal that time series-based methods are rarely employed in the detection of Tor network traffic. In this study, the performance of a long short-term memory (LSTM)-based deep learning model—commonly used in time series analysis—is investigated for the classification of Tor and NonTor network traffic alongside traditional machine learning techniques. This approach, which is supported by existing research findings, is further implemented within a graphical user interface (GUI) developed in the MATLAB environment. This interface allows users to upload datasets, apply data filtering, and perform detection using the model of their choice. The proposed system’s accuracy and overall performance demonstrate successful outcomes compared with results reported in the literature. The scientific contribution of this study lies in the development of a time series-based, GUI-supported application that presents a comparative evaluation of LSTM and conventional ML methods. Furthermore, the system provides both theoretical and practical advancements for network traffic analysis by integrating these models into a user-friendly interface.
Understanding the probability distribution ascribed to a given dataset enables commentary on the characteristics of the underlying population and formulation of prospective inferences. If the dataset is sufficiently large, the law of large numbers or the central limit theorem can be employed to ascertain the probability distribution of the dataset. However, in instances where the sample size is relatively limited, estimating the PDF of the population becomes difficult. Furthermore, determining the PDF that best describes the available data introduces an additional level of complexity to the analysis. Failure to consider this complexity can have significant consequences. This paper addresses this challenge by exploring the use of maximum entropy and Bayesian logical inference. The likelihood function is obtained with maximum entropy, and a priori distributions of means and standard deviations are assigned. The posterior distribution functions are constructed using Bayesian logical inference. The distributions derived theoretically were then applied to the milk yields of seven dairy cows. Subsequently, the performance of the Bayesian maximum entropy method (BME) is compared with that of the existing bootstrap method. The simulation results demonstrate the effectiveness, flexibility, and robustness of the BME method in obtaining the mean and variance distributions of the data when the number of data is small.
In the literature, various methods are suggested and various studies are conducted to improve existing methods to obtain the best forecasting. The purpose of this study is to show that although fuzzy time series has the disadvantage of interval selection, it can give as good results as other popular methods and emphasize that it is a more dynamic method. In this study, the monthly dam occupancy rates of the Adana/Turkey province were used as the dataset, and forecasts were made using autoregressive integrated moving average, decomposition, and fuzzy time series methods. The effect of different interval selections on the forecasting performance in fuzzy time series is also investigated. This study shows that choosing the correct interval significantly impacts the performance of fuzzy time series. This study shows that choosing the correct interval significantly impacts the performance of fuzzy time series. The prediction performance of the fuzzy time series model for extreme values was much better than other prediction methods used in the study. According to the results and evaluations, the fuzzy time series method produces dynamic and robust forecasts close to reality and gives good results with the right interval number and length selection.
The Internet of Things (IoT) has become a cornerstone of modern technology in the age of rapid digi talization, connecting billions of resourceconstrained wireless devices. While this connectivity enables numerous innovations, it also introduces serious security challenges, particularly spoofing attacks that exploitthe limited processing capabilities of IoT devices. This study proposes a novel blockchainbased model to enhance IoT network security, with a specific focus on detecting and mitigating spoofing threats. The model demonstrates strong defense capabilities by integrating the Ethereum blockchain and smart contracts into an IoT simulation environment using Raspberry Pi and Raspbian OS. The experimental setup involved simulating spoofing scenarios and measuring system performance using detection accu racy and processing time metrics. The results showed that the proposed model successfully prevented nearly 80% of spoofing attempts. However, blockchain integration led to a 42% increase in processing time, signaling the need for future optimization. Despite this tradeoff, the proposed model offers a practical and scalable solution for securing IoTsystems. This research contributes to the advancement of decentralized IoT security and sets the groundwork for future studies focusing on reducingcomputational overhead and expanding applicability across diverse IoT environments.
AI-generated images have amplified the need for effective methods to distinguish between real and synthetic visuals. This underscores the need to develop new approaches to ensure data integrity and combat misinformation. While the existing literature predominantly focuses on Generative Adversarial Networks (GAN)-based synthetic images, researchers have largely overlooked the detection of diffusionbased models. This study fills this gap by demonstrating the potential of convolutional block attention module (CBAM)-enhanced convolutional neural networks (CNNs) for the effective detection of diffusionbased synthetic images. In this study, we use CNNs enhanced with the CBAM to propose a novel approach for detecting synthetic images. The CBAM-enhanced model, trained on the CIFAKE dataset, achieved a remarkable accuracy of97.38% in detecting synthetic images. We integrate pre-trained CNN architectures, such as ResNet50 and DenseNet121, with a CBAM attention mechanism, which enhances performance by focusing on salient spatial and channel information. This approach presents a model that significantly enhances the detection capabilities for distinguishing fake images. Ourfindings contribute to the field of deepfake detection by providing a robust solution for automated digital image vetting, with implications for AI ethics, security, and broader societal discourse. The implementation details and source code are available at https://github.com/cmpe-dev/Fake-Detector-with-CBAM.
In this study, a machine learning model was developed to predict which of the seven Turkish dialects a given speech recording, collected from seven different regions of T & uuml;rkiye, belongs to. The datasets used for machine learning were gathered from YouTube, prioritizing sound recordings with high potential to reflect regional dialects, focusing on natural conversations and local people's speech. A total of15,889 s of audio was collected, ensuring a balanced representation of each regional dialect. The features of the audio recordings, segmented into specific sizes, were extracted using MFCCs. Machine learning models were then constructed with these extracted features using 11 classification methods. With the performance enhancements obtained through optimization, the classification success for each regional dialect reached an accuracy rate of 89%, while the correct prediction rates for each of the seven regions had F1-scores ranging from 80.1% to 96.6%. The results of the analysis indicate that audio recordings from the Eastern Anatolia Region were correctly predicted at a high rate of 96.6%. This study aimed to develop a machine learning model that achieves performance improvements in identifying and predicting which regional dialects audio recordings, comprising speeches with local and regional characteristic traces in T & uuml;rkiye, belong to.
This study investigates the possibilitiesofapplyingclassicalMCDM methods to grey numbers byleveraging operator overloading and multiple dispatch features of Julia programming language. Grey versions ofthe classical MCDM methods differ in some oftheir computational steps. In this paper, it is shown that a set of MCDM methods, including TOPSIS, ARAS, WASPAS, and EDAS are directly applicable to grey numbers through operator overloading without changing their core algorithms. These processes are implemented by re-defining arithmetic and comparison operators for grey numbers. We also compared our methodology with two existing Grey TOPSIS methods. It is shown that the sorted grey scores can yield the same rankings as the existing methods in some cases. When the results are not the same, we show that an appropriate whitening parameter can be found to obtain the same rankings. This approach preserves the uncertainty from the very beginning to the end of the process and provides a clear distinction between objective computational steps and subjective ranking steps. Switchingthe whitening parameterfrom zero to one also provides a sensitivity analysis tool to investigate how uncertainty affects the final rankings
Background/Purpose: There are challenges in applying artificial intelligence (AI) for acute wound healing assessment due to limited data availability, the dynamic nature of healing, and inter-patient variability. These challenges require further research on data collection, algorithm design, and validation techniques specific to each wound type. This study proposes a computer vision model for identifying acute surgical wounds using image segmentation techniques. Methods: The ACW.ai model was used to segment wound images and assign various labels to different wound features, mimicking a clinician's perspective. The model was trained and validated using ACW.ai on a chronic wound dataset (FUSeg). Results: The ACW.ai model achieved a DICE score of 90.9%, outperforming existing methods. Additionally, the model was trained on a private dataset of 59 acute wound images, achieving a DICE score of 87.90% and a mean average precision of 82.10%. Conclusion: Image segmentation is crucial for the early detection of surgical site infections (SSIs) and facilitates better assessment of wound healing. Healthcare professionals may identify potential signs of infection and monitor healing progress by segmenting wounds and analyzing specific features, enabling them to take preventive actions and improve patient outcomes.
