Sürdürülebilir Kalkınma Amaçları
Yazarlar
- Günen, Mehmet Akif(4)
- Melek, Mesut(4)
- Ozturk, Samet(3)
- Bowler, Alexander(2)
- Cam, Handan(2)
- Demirel, Ugur(2)
- Gökrem, Levent(2)
- Gül, Fatih(2)
- Reis, Hatice Catal(2)
- Unlu, Ramazan(2)
- Watson, Nicholas J.(2)
- Abdel-Fattah, Abdul-Rahman(1)
- Abdelkader, Salim Bourchid(1)
- Ahmad Khan, Khatib(1)
- Ahmed, Sana(1)
- Akbulut Yıldız, Gizem(1)
- Alemdag, Selcuk(1)
- Almasani, Heba(1)
- Arici, Hasan Evrim(1)
- Arslan, Adem(1)
- Avdan, Ugur(1)
- Ayaydın, Hasan(1)
- Aydin, Sefa(1)
- Aydin, Tolga(1)
- Aydın, Sefa(1)
- Aykas, Didem-Peren(1)
- Bacak, Aykut(1)
- Barut, Abdulkadir(1)
- Bayrakdaroglu, Serdar(1)
- Bayramoğlu, Adil(1)
- Belhaouari, Samir Brahim(1)
- Bilgilioglu, Burhan Baha(1)
- Bilgilioğlu, Burhan Baha(1)
- Cakir, Mehmet Ali(1)
- Cakir, Sümeyra Cay(1)
- Cam, Alper Veli(1)
- Catal Reis, Hatice(1)
- Celen, Ali(1)
- Celik, Ahmet(1)
- Celik, Necati(1)
Yayınlar
- Yüksek Lisans TeziAçık Erişim2026Su gövdelerindeki değişimlerin otomatik olarak tespit edilmesi ve analizi
Ozan Kiyat
Uzaktan algılama tarım, orman, bitki örtüsü ve su gibi farklı alanlarda sıklıkla kullanılmaktadır. Göllerde su yüzeyindeki değişimler, ekosistemler ve sürdürülebilir su kaynaklarının yönetimi açısından elzemdir. Bu çalışma, uydu görüntüleri kullanılarak Türkiye’de farklı oluşum kökenlerine ve su kimyalarına sahip 10 gölün (Akşehir, Eber, Tuz, Salda, Manyas, İznik, Van, Beyşehir, Eğirdir, Burdur) 1984–2025 yılları arasındaki göl su yüzey alanı değişimini otomatik olarak tespit etmeyi amaçlamaktadır. Çalışmanın temel veri kaynağını Landsat 5 TM, Landsat 8 OLI/TIRS, Landsat 9 OLI-2/TIRS-2 ve Sentinel-2 MSI uydu görüntüleri oluşturmaktadır. CHIRPS yağış verileri ile ERA5-Land potansiyel buharlaşma verileri hidroklimatik değerlendirmeleri desteklemek amacıyla kullanılmıştır. Veriler, Google Earth Engine platformunda işlenmiştir. Su yüzeylerinin sınıflandırılmasında spektral indeksler ve Otsu eşikleme yöntemi kullanılmış; bu kapsamda Modifiye Edilmiş Normalize Edilmiş Fark Su İndeksi (MNDWI), Normalize Fark Bitki Örtüsü İndeksi (NDVI), Normalize Fark Su İndeksi (NDWI) ve Otomatik Su Çıkarma İndeksi (AWEI) hesaplanmıştır. Ayrıca tüm analiz yıllarında (1984, 1990, 1995, 2000, 2005, 2010, 2015, 2020, 2025) Rastgele Orman (RO), Destek Vektör Makineleri (DVM), Sınıflandırma ve Regresyon Ağacı (CART) ve K-En Yakın Komşu (KNN) makine öğrenmesi algoritmaları uygulanmıştır. Göllerin su yüzey alanlarında anlamlı azalmalar tespit edilmiştir. En çarpıcı düşüşler, yüzey alanının tamamını yitiren (%100) Akşehir Gölü’nde ve alanının %93,68’ini kaybeden Tuz Gölü’nde gözlemlenmiştir. Makine öğrenmesi algoritmaları arasında RO modeli, %99,25 genel doğrulukla en yüksek sınıflandırma başarısını göstermiştir. Bu çalışma, farklı oluşum ve özelliklere sahip göllerin izlenmesi için genelleştirilebilir ve güncellenebilir bir uzaktan algılama iş akışı sunmaktadır. Ayrıca sürdürülebilir su kaynakları yönetimi için farklı disiplinlere bilimsel bir altlık sağlamaktadır.
- Yüksek Lisans TeziAçık Erişim2026Çayeli (Rize) ilçesinde jeofizik verilerle bütünleşik heyelan duyarlılığı ve risk haritalarının oluşturulması
Zehra Sürül
Bu çalışmada, Çayeli ilçesinde heyelan duyarlılığının belirlenmesi amacıyla makine öğrenmesi tabanlı bütünleşik bir analiz gerçekleştirilmiştir. Çalışmada topoğrafik, hidrolojik, jeolojik, iklimsel ve sismik/jeoteknik faktörler birlikte değerlendirilmiş; AFAD, MTA, CORINE, OpenStreetMap ve sayısal yükseklik modeli verileri kullanılmıştır. Ayrıca Çayeli Belediyesi’nden temin edilen 55 adet jeolojik-jeoteknik rapor incelenmiş ve jeofizik veriler CBS ortamında modele dahil edilmiştir. Toplam 28 koşullayıcı faktör oluşturulmuş, VIF analizi sonucunda yüksek doğrusal bağlantıya sahip değişkenler çıkarılarak modelleme süreci 22 faktör üzerinden yürütülmüştür. Lojistik Regresyon, Rastgele Orman, XGBoost, LightGBM, SVM ve YSA dahil olmak üzere on farklı makine öğrenmesi algoritması karşılaştırılmıştır. Modeller 5 katlı çapraz doğrulama ve bağımsız test veri seti ile değerlendirilmiştir. Son aşamada SHAP analizi uygulanarak faktörlerin model tahminlerine etkisi yorumlanmıştır. Elde edilen sonuçlar doğrultusunda heyelan duyarlılık haritaları üretilmiş ve çalışma alanındaki mevcut heyelanlarla yüksek mekânsal uyum gösterdiği belirlenmiştir. Çalışma, makine öğrenmesi ve açıklanabilir yapay zekâ yöntemlerinin heyelan duyarlılık analizlerinde etkili sonuçlar sunduğunu ortaya koymuştur.
- MakaleYalnız künye2026Classification and detection of Salmonella, Escherichia coli O157:H7, and Listeria monocytogenes using Fourier-transform near infrared spectroscopy coupled with machine learning
Ozturk, Samet; Huang, Lihan; Hwang, Cheng-An; Sheen, Shiowshuh
This study was conducted to investigate the potential use of Fourier-transform near-infrared (FT-NIR) spectroscopy combined with machine learning (ML) algorithms to accurately identify three foodborne pathogens, including Salmonella spp., Listeria monocytogenes, and Escherichia coli O157:H7. Each bacterial strain (two per pathogen) was individually cultured and purified through sequential washing using ethanol–deionized (DI) water solutions. Each purified culture was transferred into each well of a 96-well cell plate covered with custom-cut filter paper and then vacuum dried at 50 °C under 20 kPa for 1 h. The absorbance spectra of dehydrated bacterial cells were then acquired across the range of 1000–2400 nm using a diffuse reflectance probe connected to a FT-NIR process analyzer. To identify the optimal classification pipeline, the acquired spectra were analyzed using ten different pre-processing methods, three feature selection methods, and supervised algorithms including partial least squares discriminant analysis (PLS-DA), support vector machine (SVM), random forest (RF), artificial neural network (ANN), and convolutional neural network (CNN). Furthermore, the accuracy of the developed ML pipelines was further enhanced through under-sampling and boosting steps. The results demonstrated that SVM, RF, ANN and CNN outperformed PLS-DA, where the classification accuracies were above 90%. The findings demonstrate that using Savitzky-Golay first derivative (SG1) filtering to pre-process the full spectra followed by SVM classification yielded the highest accuracy, achieving 95.3% overall accuracy in ML pipeline. This study highlights the powerful capability of FT-NIR spectroscopy coupled with ML algorithms to detect and identify foodborne pathogens on dehydrated surfaces in process environments.
- MakaleYalnız künye2026Machine learning-based multi-class classification of rock mass permeability using integrated geotechnical and geophysical parameters
Sari, Mahmut; Alemdag, Selcuk; Melek, Mesut; Seren, Aysel
In engineering applications, machine learning (ML) provides a promising approach when direct characterization of subsurface properties is challenging, time-consuming, and costly. This study developed a multi-class ML framework for classifying rock mass permeability based on Lugeon values using 86 observations from drilling and geophysical investigations at the Kırklartepe dam site in Bayburt, Türkiye. Seven predictors were considered: depth, Rock Quality Designation (RQD), joint frequency, Schmidt hammer hardness, P-wave velocity, S-wave velocity, and electrical resistivity. The observations were assigned to four permeability classes: very low, low, moderate, and high. Six classifiers were evaluated: Polynomial Kernel Support Vector Machine (P-SVM), Linear Discriminant Analysis, k-Nearest Neighbors (k-NN), AdaBoostM2, Bagged Trees, and Radial Basis Function SVM. Sequential Forward Feature Selection and Leave-One-Out Cross-Validation were used for feature selection and model evaluation, respectively. Robustness was further assessed using bootstrap confidence intervals, class-imbalance treatments, feature-selection stability, and learning curves. Among the baseline models, P-SVM using all seven predictors achieved the highest accuracy (79.07%), with a macro-F1 of 0.774 and the lowest Golden Distance (GD = 0.4372). When balanced class-wise performance was prioritized, k-NN with in-fold SMOTE achieved the lowest overall GD (0.4126) and a macro-F1 of 0.777. S-wave velocity and electrical resistivity were the most influential predictors. Overall, the findings demonstrate that the proposed framework effectively integrates geotechnical and geophysical data for rock mass permeability classification, providing a robust, data-driven decision-support approach that supports subsurface characterization and complements conventional permeability assessment in engineering applications.
- MakaleAçık Erişim2026A leakage-free analytics framework for subject-wise electrocardiogram-based sleep apnea detection using machine learning
Melek, Negin
Sleep apnea is associated with impaired sleep quality and elevated cardiovascular risk, creating a need for reliable automated screening approaches. Although electrocardiogram (ECG)-based detection methods have shown promising results, reported performance values can be affected by evaluation strategies that allow information leakage between training and testing data. This study investigates ECG-based sleep apnea detection under strict subject-wise conditions using the PhysioNet Apnea-ECG database. Thirty-five annotated single-lead ECG recordings sampled at 100 Hz were divided into non-overlapping 60-s segments according to minute-wise apnea labels. Statistical ECG descriptors and time-domain heart rate variability features were extracted and evaluated using k-nearest neighbors (KNN), support vector machine (SVM), and Random Forest (RF) classifiers. In addition, a lightweight one-dimensional convolutional neural network (1D-CNN) was trained directly on filtered ECG segments. Repeated holdout validation with subject-level train-test separation was adopted, and normalization parameters were estimated exclusively from training data in each repetition to prevent information leakage. Model performance was assessed using accuracy, balanced accuracy, precision, recall, specificity, F1-score, ROC-AUC, Cohen's kappa, and Golden Distance (GD). In the representative subject-wise evaluation, Random Forest achieved the most balanced performance, yielding the highest balanced accuracy (0.70) together with the lowest GD (0.56). Across the repeated subject-wise holdout experiments, SVM achieved the strongest average overall performance, producing the highest mean accuracy (0.72), balanced accuracy (0.69), ROC-AUC (0.77), and Cohen's kappa (0.39), and the lowest mean GD value (0.58). The lightweight 1D-CNN converged stably during training, but its performance on unseen subjects remained lower than that of the handcrafted feature-based models.
- MakaleAçık Erişim2026Real maritime measurement-driven machine learning for path loss modeling in shore-to-ship wireless links
Sağır, Nuğman; Eyüboğlu, Ekrem; Kuzulugil, Kenan; Tuğcu, Zeynep Hasırcı; Çavdar, İsmail Hakkı
Reliable propagation modeling remains a fundamental requirement for the design and deployment of maritime wireless communication systems, particularly in shore-to-ship scenarios characterized by highly variable environmental conditions. Unlike conventional empirical and deterministic models that rely on theoretical assumptions, the proposed approach is entirely measurement-driven, enabling more accurate and robust propagation modeling. Four regression algorithms—Gaussian Process Regression, Least Squares Boosting, Decision Tree, and Support Vector Regression—are evaluated using a feature set that includes transmission distance, sea state, weather conditions, and antenna configurations. Among the tested models, the Decision Tree achieves the best performance, with an RMSE of 1.18 dB and an R2 of 0.96, corresponding to an error reduction of approximately 45.6% compared to the best-performing empirical model. Feature importance analysis shows that transmission distance and antenna geometry dominate signal attenuation, while environmental variables have a limited influence at this frequency. Unlike most existing studies that emphasize deep learning or black-box models, this work demonstrates that lightweight and interpretable regression models can achieve comparable accuracy while enabling real-time link planning and practical deployment. The focus is on engineering applicability rather than proposing new learning architectures. Overall, the results indicate that measurement-driven machine learning provides accurate and interpretable solutions for shore-to-ship path loss modeling under real maritime conditions.
- MakaleAçık Erişim2026Enhancing precision of low-cost detectors via machine learning for accurate mass attenuation coefficient estimation
Düzgün, İ.; Küçük, M.A.; Kaya, S.; Çelik, N.
Low-cost NaI(Tl) detectors are practical for gamma-ray spectrometry but their limited energy resolution can introduce systematic errors in mass attenuation coefficient measurements when scattered photons are included in the peak region. This study proposes a machine-learning correction framework that maps NaI(Tl)-derived attenuation coefficients to XCOM-based HPGe-equivalent reference values through a dimensionless correction factor. An EGS4/XCOM dataset of 756 model-development configurations, covering 18 atomic numbers, six monoenergetic photon energies, and seven collimator radii, was used to train twelve regression models and evaluate them over 60 randomized train–validation–test cycles. Sodium and Bismuth were withheld for external material-wise validation. Gaussian Process Regression with a Matérn 5/2 kernel provided the best overall testing performance. For the withheld materials, the correction reduced the mean deviation from XCOM references from 3.07% to 0.83%, supporting its use as a practical correction layer for low-cost attenuation measurements.
- MakaleYalnız künye2026Comparison of Zn recovery prediction from carbonate ores with machine-learning methods
Erkan, Ilker; Günen, Mehmet Akif
This study addresses the challenge of predicting zinc (Zn) recovery from carbonate ores via sodium hydroxide (NaOH) leaching. This complex process influenced by variable ore composition, surface passivation effects, and nonlinear reaction dynamics, which complicate reagent optimization and process control in hydrometallurgical operations. To tackle this, a dataset containing 422 experimental observations was compiled from previous studies, incorporating ore composition and process parameters, such as NaOH concentration, leaching time, temperature, and solid-to-liquid ratio. Four regression models (decision tree, neural network, generalized additive model, and random forest) were trained and evaluated using performance metrics, such as coefficient of determination (R2), root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and symmetrical mean absolute percentage error (SMAPE). Among these, the random forest model achieved the best predictive accuracy, with R2 value of 0.8541 on the test set and the lowest error rates, demonstrating its effectiveness in capturing the complex relationships between input variables and Zn recovery. Explainable artificial intelligence, particularly SHapley additive exPlanations (SHAP) analysis, revealed that NaOH concentration, leaching time, and solid-to-liquid ratio had the most positive influence on Zn recovery, whereas elements such as Ca, Fe, and Pb had inhibitory effects. These findings align with known geochemical behavior and provide valuable insights for reagent optimization and process efficiency in leaching processes. This study demonstrates the practical potential of machine learning in mineral processing, offering a scalable framework for optimizing Zn recovery from non-sulfide ores and a data-driven approach to enhance decision-making in hydrometallurgical applications.
- MakaleAçık Erişim2026Towards Greening the BRICS: Uncovering the Impact of Green Energy, Green Technology and Forest Cover on Environmental Quality
Henni, Mohamed Djafar; Ayaydın, Hasan; Akbulut Yıldız, Gizem; Orhan, Abdullah; Pilatin, Abdulmuttalip; Abdelkader, Salim Bourchid
The BRICS economies, facing the threat of climate change, face a policy challenge in transitioning from fossil-fuel-based energy systems and improving environmental quality. This necessitates urgent policy changes in the outdated energy infrastructure of BRICS countries. However, there still remains a policy gap regarding how countries in the BRICS, a group of rapidly developing economies, can grow their economies in line with the Sustainable Development Goals. The aim of our study is to investigate the impact of green energy, green technology, and forest cover on environmental quality in BRICS countries. The BRICS group of countries offers an ideal field of study for both examining the impacts of green energy, green technology, and forest areas on environmental quality in developing economies, as well as for evaluating national and global energy policies. Although numerous studies have empirically examined the relationship between environmental variables and green initiatives, the impact of green initiatives on the load capacity factor has been overlooked. These studies have generally used various econometric methods and have not included machine learning in the process. This study addresses this research gap by focusing on green energy and green technology, which are preferred for their various advantages and make significant contributions to the load capacity factor. To analyze this relationship in BRICS economies between 2000 and 2022, the Augmented Mean Group (AMG) estimator and Machine Learning algorithms were used. According to the results, strong evidence emerges of a positive relationship between green energy, the digital economy, forested area, and the load capacity factor, while a negative relationship exists between green technology, growth, and the load capacity factor. Based on robust empirical findings, renewable energy sources are a key driver of clean energy adoption and can ultimately increase the load capacity factor in BRICS economies. The results also imply that, since developments in green technological innovation in BRICS countries are still in their infancy, investments in green technologies for a sustainable environment need to be qualitatively increased.
- MakaleAçık Erişim2026Land Subsidence Susceptibility Mapping in Semi-Arid Regions Using PS-InSAR and Tree-Based Ensemble Machine Learning: A Case Study From Çumra, Türkiye
Bilgilioğlu, Burhan Baha
Land subsidence is an increasing environmental hazard in semi-arid agricultural basins where intensive groundwater abstraction, compressible geological units, and expanding land-use pressures interact. This study presents a PS-InSAR and machine-learning-based framework for land subsidence susceptibility mapping in the Çumra District of the Konya Closed Basin, Türkiye. A time-series PS-InSAR analysis was conducted using 172 Sentinel-1A ascending images acquired between 2018 and 2023, and 11,718 high-confidence subsidence points were extracted as the inventory dataset. Fourteen conditioning factors representing topographic, hydrogeological, lithological, environmental, and anthropogenic controls were integrated with eight tree-based ensemble models optimized using Optuna. Random Forest achieved the highest overall performance, with an accuracy of 0.9745, precision of 0.9562, recall of 0.9945, F1 score of 0.9750, and Cohen's Kappa of 0.9490, followed by Extremely Randomized Trees and XGBoost. Class-specific validation also confirmed the reliability of Random Forest in delineating the Very High Susceptibility class. Variable importance analysis identified land use, proximity to settlements, well density, and groundwater level change as the dominant predictors, emphasizing the role of groundwater-related anthropogenic pressure in subsidence development. The resulting susceptibility maps provide a spatial decision-support basis for groundwater management, land use planning, and subsidence risk mitigation.
- Derleme MakaleAçık Erişim2026Artificial-intelligence–based developmental screening for pre-school children (ages 3–6): a scoping review of deployed tools, clinical validity, equity, and implementation in guidance-center settings
Arslan, Adem; Gül, Fatih
Background – Early identification of neurodevelopmental conditions during the pre-school years (ages 3–6) is critical for timely intervention, yet traditional questionnaire-based screeners suffer from low positive predictive value, late detection, and inequitable access. Artificial intelligence (AI) and machine-learning (ML) methods promise more objective, scalable identification, and two devices (Cognoa Canvas Dx, 2021; EarliPoint Evaluation, 2023) are already US Food and Drug Administration (FDA)-authorized. Recent reviews summarize the technical literature, but few address how deployed tools perform on clinical validity, equity, and real-world implementation, particularly outside high-income, English-speaking settings. Methods – Following Joanna Briggs Institute methodology and the PRISMA extension for scoping reviews (PRISMA-ScR), we searched seven databases (1 January 2020–15 June 2026) and charted primary validation studies of AI/ML developmental-screening tools for children aged 72 months or younger. We mapped tools by modality, diagnostic performance, regulatory status, geographic origin, and, as a structured analytic axis, demographic subgroup (equity) reporting and deployment readiness. Results – Sixteen primary AI-tool studies met inclusion, spanning eight modality classes (eye-tracking, multimodal questionnaire-plus-video, computer-vision home/clinical video, tablet/smartphone digital phenotyping, voice/acoustic ML, electronic-health-record risk prediction, questionnaire-based ML, and wearable sensors) and eight countries (USA, South Korea, Australia, Switzerland, France, Israel, China, Vietnam). Reported sensitivities ranged from approximately 75 to 99.2% and specificities from 78.9 to 95.5%. Only one tool demonstrated stable performance across sex, race, and ethnicity; subgroup-stratified validation was otherwise near-absent. No CE-marked autism-specific screening device was identified. Conclusions – The evidence base has broadened in modality and geography but remains autism-concentrated and equity-blind, with a wide gap between analytic accuracy and primary-care deployment. We propose equity validation, deployment-readiness reporting, and an implementation pathway for low-resource guidance-center settings as priorities.
- MakaleAçık Erişim2026Predictors of citations: an analysis of highly-cited-papers in hospitality and tourism research using a machine learning approach
Polat, Eray; Çelik, Fatih; Arici, Hasan Evrim; Köseoglu, Mehmet Ali
In the dynamic nature of hospitality and tourism (H&T) research, it is increasingly difficult to distinguish highly-cited-papers (HCPs) due to the rapid proliferation of publications. This study employs machine learning techniques to identify the predictors of citation counts in H&T research over both short-term (5-year) and long-term (20-year) periods using HCPs. The analysis integrates a theoretical framework comprising normative theory and social constructivist theory. The findings indicate that international citation, PlumXmetrics, and early citations are the most effective determinants in both periods. Furthermore, while the importance of international citations is evident in both periods, the order of importance of the other two predictors changes. PlumXmetrics are more important in the long-term, while early citations are more important in the short-term. In conclusion, this comprehensive and up-to-date study of citation dynamics provides valuable insights for scholars and other stakeholders interested in enhancing the visibility and influence of H&T literature.
- MakaleAçık Erişim2026Assessing the Predictive Value of Kolmogorov–Arnold Networks for the No–Reflow Phenomenon in ST–Segment Elevation Myocardial Infarction: A Comparative Machine Learning Study; [ST Segment Yükselmeli Miyokard Enfarktüsünde No–Reflow Fenomeni için Kolmogorov–Arnold Ağlarının Öngörü Değerinin Değerlendirilmesi: Karşılaştırmalı Makine Öğrenimi Çalışması]
Taşolar, Hakan; Bayramoğlu, Adil; Günen, Mehmet Akif; Levent, Sümeyye; Güral, Yunus; Halisdemir, Nurhan
Objective: The no-reflow phenomenon in ST-segment elevation myocardial infarction (STEMI) is a significant clinical issue associated with poor cardiovascular outcomes. This study aimed to develop and compare multiple supervised machine learning algorithms, including the recently introduced Kolmogorov-Arnold Network (KAN), to predict the occurrence of the no-reflow phenomenon in patients with STEMI undergoing primary percutaneous coronary intervention (PCI). Method: This prospective, single-center study included 890 consecutive STEMI patients undergoing primary PCI. The Synthetic Minority Over-sampling Technique (SMOTE) was utilized to address class imbalance during training. Feature selection using analysis of variance (ANOVA) F-statistics and validation of feature independence (Variance Inflation Factor [VIF] < 5) identified ejection fraction (EF), baseline troponin level, stent length, B-type natriuretic peptide (BNP) level, and total ischemic time as the most influential predictors. Results: The KAN and Extreme Gradient Boosting (XGBoost) models achieved the highest predictive accuracy (area under the curve > 0.98, F1 > 0.95), outperforming traditional models such as logistic regression and decision tree classifiers (DeLong test, P < 0.001). Feature selection improved efficiency and reduced runtime by 20–40%, while Shapley Additive exPlanations-based (SHAP-based) explainability confirmed that the predictions were physiologically consistent: higher EF and lower BNP reduced the probability of no-reflow, whereas longer stent length and ischemic time increased it. The superior performance of KAN and XGBoost underscores the importance of modeling nonlinear relationships and multidimensional interactions among clinical, laboratory, and procedural variables. Conclusion: These findings suggest that KAN may serve as a reliable analytical framework for exploring complex cardiovascular outcomes. However, further multicenter and externally validated studies are needed to confirm its generalizability and potential role in clinical risk assessment.
- MakaleAçık Erişim2026YOLOv8 computer vision for automated offside detection in professional football validated through supervised learning
Abdel-Fattah, Abdul-Rahman; Belhaouari, Samir Brahim; Ceylan, Halil Ibrahim; Bayrakdaroglu, Serdar; Muntean, Raul Ioan; Dhahbi, Wissem; Dergaa, Ismail; Omri, Abdelfatteh El
Offside calls are integral to maintaining fairness and competitive integrity in football, yet current officiating processes are constrained by limitations in accuracy despite technological advancements. This investigation developed and validated an automated offside detection system using YOLOv8 object detection architecture to address documented limitations of Video Assistant Referee (VAR) systems in offside decision accuracy. Supervised learning validation employed 400 annotated instances from professional match footage. The YOLOv8 medium variant was configured for four-class detection, with a focus on extreme player positions, reducing computational complexity by approximately 70% relative to comprehensive player tracking while accepting accuracy trade-offs in marginal scenarios. The system was optimized for computational efficiency through highly accurate tracking of player positions. Data were partitioned into 40% (n = 160) for training and 60% (n = 240) for testing, with 5-fold cross-validation. The automated system achieved robust performance metrics across 240 test instances: overall accuracy of 83.0%, precision of 85.0%, recall of 87.0%, and F1-score of 86.0%. Confusion matrix analysis revealed true positives (n = 130), true negatives (n = 70), false positives (n = 22), and false negatives (n = 18). Individual case analysis demonstrated 80% agreement with ground truth classifications. System performance significantly exceeded the random classification baseline (p < 0.001). The strategic extreme player tracking approach achieved substantial computational efficiency while maintaining decision accuracy across diverse match scenarios. A YOLOv8-based offside detection system demonstrates proof-of-concept feasibility for decision-support applications in football officiating. Performance (83% accuracy, 95% CI: 79-87%) on 240 test instances remains below professional benchmarks (> 93%), with single-camera constraints and limited dataset size restricting generalizability to proof-of-concept validation rather than operational deployment.
- MakaleAçık Erişim2025Investigation of Fluctuations in Cryptocurrency Transactions with Sentiment Analysis
Demírel, Ugur; Çam, Handan
This study investigates public sentiment about popular cryptocurrencies listed on crypto exchanges in Turkey, using comments shared on social media platforms and online forums. The research seeks to enhance the existing body of knowledge by overcoming the shortcomings of sentiment analysis studies focused on Turkish texts. Data collected from social media and online forums were examined with sentiment analysis techniques. A total of 607,592 comments were analyzed, of which 89,986 were classified as negative, 72,655 as positive, and 444,951 as neutral. For binary classification, 89,986 negative and 72,655 positive examples were selected and machine-learning models were trained and tested on 162,641 examples. The study's methodology includes an in-depth examination of sentiment analysis results obtained using machine learning classifiers. The findings show how various cryptocurrencies are perceived on different social media platforms. For instance, BTC (Bitcoin) is generally perceived negatively on Investing.com and Telegram, while ETH (Ethereum) generally displays more negative views. These results help investors understand their perceptions and market expectations towards cryptocurrencies. This study deepens the role of social media sentiment analysis in cryptocurrency markets, contributing to the development of new methods and approaches for future research.
- MakaleYalnız künye2025Interpretable machine learning for rapid prediction and calibration of HPGe detector efficiency: a physics-informed approach and online platform
Yesilkanat, Cafer Mert; Celik, Necati; Celik, Ahmet; Cevik, Ugur
This study introduces a novel, physics-informed, and calibration-friendly hybrid machine learning framework for the rapid and accurate prediction of the Full Energy Peak (FEP) efficiency in High-Purity Germanium (HPGe) detectors. To overcome the limitations of conventional “black-box” models, our two-stage approach first represents the FEP efficiency curve using a physically interpretable logarithmic polynomial. Subsequently, machine learning models were trained to predict the polynomial coefficients directly from the detector geometric parameters using a comprehensive dataset generated via Monte Carlo simulations. Among the various algorithms tested, the Generalized Linear Model yielded superior performance, achieving R2 values of 0.975–0.992 for the coefficients. While raw model predictions showed expected variability, a key feature of our framework (a single-point calibration protocol using one known efficiency value) dramatically improved accuracy, reducing the mean absolute percentage error by an average of 80% on the independent test data. The calibrated model was further validated using an experimental detector. The entire framework was deployed as a user-friendly online platform, enabling researchers to instantly generate and calibrate efficiency curves. Our work successfully harmonizes interpretability, speed, and accuracy, offering a powerful tool for the design, optimization, and routine calibration of HPGe detectors.
- MakaleAçık Erişim2025Intersession Robust Hybrid Brain–Computer Interface: Safe and User-Friendly Approach with LED Activation Mechanism
Aydın, Sefa; Melek, Mesut; Gökrem, Levent
This study introduces a hybrid Brain–Computer (BCI) system with a robust and secure activation mechanism between sessions, aiming to minimize the negative effects of visual stimulus-based BCI systems on user eye health. The system is based on the integration of Electroencephalography (EEG) signals and Electrooculography (EOG) artefacts, and includes an LED stimulus operating at a frequency of 7 Hz for safe activation and objects moving in different directions. While the LED functions as an activation switch that reduces visual fatigue caused by traditional visual stimuli, moving objects provide command generation depending on the user’s intention. In order to evaluate the stability of the system against physiological and psychological conditions, data were collected from 15 participants in two different sessions. The Correlation Alignment (CORAL) method was applied to the data to reduce the variance between sessions and to increase stability. A Bootstrap Aggregating algorithm was used in the classification processes, and with the CORAL method, the system accuracy rate was increased from 81.54% to 94.29%. Compared to similar BCI approaches, the proposed system offers a safe activation mechanism that effectively adapts to users’ changing cognitive states throughout the day by reducing visual fatigue, despite using a low number of EEG channels, and demonstrates its practicality and effectiveness by performing on par or superior to other systems in terms of high accuracy and robust stability.
- BildiriYalnız künye2025Improving Skin Lesion Classification with Pretrained Deep Learning Models
Almasani, Heba; Ture, Hayati
Skin cancer is the most common type of cancer. Cancer can be defined as a disease characterized by the uncontrolled division and spread of cells. As with other types of cancer, early diagnosis and treatment play a crucial role in combating skin cancer. Since diagnosis is made through visual examination, factors such as the level of expertise and image artifacts negatively affect diagnostic accuracy and slow down the process.Computer-aided applications used to analyze such images provide faster and more accurate results. Additionally, these methods eliminate factors that reduce diagnostic accuracy during the classification and segmentation of skin lesions.This study classifies skin cancer to benign or malignant using deep learning and machine learning methods. In particular, Convolutional Neural Networks (CNNs) have shown promising results in medical image classification. The study evaluates six pre-trained CNN models - VGG16, EfficientNetB7, EfficientNetV2L, DenseNet, ConvNeXtBase and MobileNet - for binary skin lesion classification using the ISIC 2019 dataset. The study first preprocesses and balances the data to enhance performance. The results indicate that the ConvNeXtBase model achieved the highest success with accuracy rate of 84.63% after balancing the data that we have.
- MakaleYayıncıda açık erişim2025Explainable artificial intelligence for rock discontinuity detection from point cloud with ensemble methods
Günen, Mehmet Akif
This study presents a framework for the semi-automatic detection of rock discontinuities using a three-dimensional (3D) point cloud (PC). The process begins by selecting an appropriate neighborhood size, a critical step for feature extraction from the PC. The effects of different neighborhood sizes (k = 5, 10, 20, 50, and 100) have been evaluated to assess their impact on classification performance. After that, 17 geometric and spatial features were extracted from the PC. Next, ensemble methods, AdaBoost.M2, random forest, and decision tree, have been compared with Artificial Neural Networks to classify the main discontinuity sets. The McNemar test indicates that the classifiers are statistically significant. The random forest classifier consistently achieves the highest performance with an accuracy exceeding 95% when using a neighborhood size of k = 100, while recall, F-score, and Cohen's Kappa also demonstrate high success. SHapley Additive exPlanations (SHAP), an Explainable AI technique, has been used to evaluate feature importance and improve the explainability of black-box machine learning models in the context of rock discontinuity classification. The analysis reveals that features such as normal vectors, verticality, and Z-values have the greatest influence on identifying main discontinuity sets, while linearity, planarity, and eigenvalues contribute less, making the model more transparent and easier to understand. After classification, individual discontinuity sets were detected using a revised DBSCAN from the main discontinuity sets. Finally, the orientation parameters of the plane fitted to each discontinuity were derived from the plane parameters obtained using the Random Sample Consensus (RANSAC). Two real-world datasets (obtained from SfM and LiDAR) and one synthetic dataset were used to validate the proposed method, which successfully identified rock discontinuities and their orientation parameters (dip angle/direction).
- MakaleYayıncıda açık erişim2025GeMID: Generalizable models for IoT device identification
Kostas, Kahraman; Yasa Kostas, Rabia; Just, Mike; Lones, Michael A.
With the proliferation of devices on the Internet of Things (IoT), ensuring their security has become paramount. Device identification (DI), which distinguishes IoT devices based on their traffic patterns, plays a crucial role in both differentiating devices and identifying vulnerable ones, closing a serious security gap. However, existing approaches to DI that build machine learning models often overlook the challenge of model generalizability across diverse network environments. In this study, we propose a novel framework to address this limitation and to evaluate the generalizability of DI models across data sets collected within different network environments. Our approach involves a two-step process: first, we develop a feature and model selection method that is more robust to generalization issues by using a genetic algorithm with external feedback and datasets from distinct environments to refine the selections. Second, the resulting DI models are then tested on further independent datasets to robustly assess their generalizability. We demonstrate the effectiveness of our method by empirically comparing it to alternatives, highlighting how fundamental limitations of commonly employed techniques such as sliding window and flow statistics limit their generalizability. Moreover, we show that statistical methods, widely used in the literature, are unreliable for device identification due to their dependence on network-specific characteristics rather than device-intrinsic properties, challenging the validity of a significant portion of existing research. Our findings advance research in IoT security and device identification, offering insight into improving model effectiveness and mitigating risks in IoT networks.
