Sürdürülebilir Kalkınma Amaçları

• 4
• 3
• 3
• 2
• 2
• 1
• 1
• 1

• MakaleAçık Erişim2027/r/gumushane/14229
  A hierarchical metaheuristic framework for CPT-based soil liquefaction prediction with Kolmogorov–Arnold Network variants

  Günen, Mehmet Akif; Öztürk, Kaşif Furkan

  Soil liquefaction is a critical geotechnical hazard that threatens the safety of civil infrastructure under seismic loading, necessitating robust and interpretable prediction models. This study proposes a hierarchical framework for CPT-based soil liquefaction prediction using meta-heuristic-optimized Kolmogorov–Arnold Network (KAN) variants. The framework follows a two-stage optimization strategy. First, six state-of-the-art meta-heuristic algorithms are statistically compared to identify the most effective optimization method. Second, the selected algorithm is employed to optimize six KAN variants, including FastKAN, EfficientKAN, FourierKAN, ChebyKAN, JacobiRKAN, and PadéRKAN. Statistical comparisons based on the Wilcoxon signed-rank test demonstrate that the Colony-Based Search algorithm (CSA) achieves lower objective-function values than the competing meta-heuristics (p<0.01) and is therefore adopted as the core optimization strategy. Performance evaluation reveals that the CSA-optimized ChebyKAN model achieves the highest predictive accuracy, with an F1 Score of 0.8875 ± 0.011 and an AUC of 0.9115 ± 0.013. Pairwise McNemar tests detected no statistically significant difference in classification decisions for any evaluated model pair (p>0.01). SHapley Additive exPlanations analysis identified qc, qc1N, and amax as the variables with the largest model-specific attributions, whereas d, c, and σv received lower attributions. A cost-sensitive error analysis further showed that ChebyKAN produced the lowest false-negative rate and the lowest asymmetric error cost under 5:1 and 10:1 false-negative-to-false-positive cost scenarios, supporting its conservative behaviour for safety-critical liquefaction screening. The results establish the proposed CSA–KAN framework as a reliable, high-accuracy, and interpretable solution for CPT-based soil liquefaction assessment in geotechnical engineering practice.

  • 
  İncele
• MakaleYalnız künye2026/r/gumushane/13804
  Very fast Kolmogorov-Arnold network for landslide susceptibility prediction

  Günen, Mehmet Akif; Tona, Aziz Uğur

  Landslides represent critical geological events causing loss of life and property. Prediction of landslide susceptibility is essential for risk reduction and decision-making. XGBoost, LightGBM, NGBoost, CatBoost, MLP, and FastKAN were used to predict landslide susceptibility in Gümüşhane, Türkiye, a region with complex topographic and geological structures. 20 factors were used: 9 topographical, 3 hydrological, 4 climatic, 2 geological, and 2 environmental and infrastructure-related factors. Model performance was assessed using Accuracy, Precision, Recall, and F-Score, with SHAP analysis showing feature contributions. A multicollinearity analysis was used to determine the final feature set. Topographical variables (Aspect, Slope) showed lower impacts, while geological and environmental factors (Lithology, Distance to Fault) demonstrated moderate influence across models. For FastKAN, DEM, precipitation, and distance to fault were key factors, while for boosting methods, GHI, Tmax, and DEM were most significant. FastKAN achieved the highest performance under the internal pixel-wise evaluation scheme, with an accuracy of 96.54%, followed by XGBoost at 95.56%. McNemar’s statistical test showed no significant difference between NGBoost and MLP. Post-inventory landslide events observed in 2025 and 2026 were spatially consistent with high and very high susceptibility zones, providing a qualitative consistency check rather than rigorous independent quantitative validation.

  İncele
• Editöre MektupAçık Erişim2026/r/gumushane/13730
  Reply to the Letter to the Editor: Assessing the Predictive Value of Kolmogorov–Arnold Networks for the No–Reflow Phenomenon in ST–Segment Elevation Myocardial Infarction: A Comparative Machine Learning Study; [Editöre Mektup Yanıtı: ST Segment Yükselmesi 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

  • 
  İncele
• MakaleYalnız künye2026/r/gumushane/13713
  Burned area detection using Sentinel-1 and Sentinel-2 features and analysis of ensemble models with explainable AI

  Günen, Mehmet Akif; Türk, Salih

  Wildfires are serious environmental disasters that threaten ecosystem sustainability, reduce biodiversity, and disrupt the global carbon cycle. This study aimed to improve the accuracy and interpretability of burned area mapping in the Bodrum, Köyceğiz, and Marmaris districts of Muğla province, Türkiye, by integrating multi-sensor satellite data with ensemble machine learning and explainable artificial intelligence (XAI). Sentinel-1 and Sentinel-2 imagery were processed to derive 50 optical and radar-based features, which were classified using four ensemble methods: AdaBoost.M1, Random Forest, GentleBoost, and RUSBoost. The SHapley Additive exPlanations framework was applied to quantify the contribution of each feature and guide feature selection. Results indicated that Sentinel-2 spectral indices, particularly NBR, NBRSWIR, and bands B12 and B8, were the most decisive variables for burned area classification, while Sentinel-1 features contributed less strongly overall but the IVH_ratio variable emerged as the most informative radar-based indicator. Among the classifiers, AdaBoost.M1 achieved the most stable and accurate performance, Random Forest showed strong sensitivity to spectral variability, GentleBoost performed effectively with reduced feature sets, and RUSBoost addressed class imbalance but showed weaker spatial continuity. Importantly, classification accuracy saturated after approximately 15 features, demonstrating that high accuracy can be achieved with fewer inputs and lower computational demand. Moreover, McNemar’s statistical test confirmed that the observed differences among classifiers were significant, highlighting GentleBoost’s clear superiority over other methods. Overall, the integration of optical and radar data with XAI provides a transparent, efficient, and replicable framework for operational burned area monitoring in Mediterranean ecosystems.

  • 
  • 
  İncele
• MakaleAçık Erişim2026/r/gumushane/13676
  Modified very deep super resolution network-Gaussian mixture models based single image super-resolution: A comprehensive comparative analysis

  Günen, Mehmet Akif

  Image Super-Resolution (SISR) is a vital, cost-effective solution for enhancing the spatial resolution of satellite imagery. However, a significant challenge in current SISR techniques is that while deep learning models excel at structural learning, they often struggle to accurately reconstruct intricate high-frequency details and preserve the statistical texture properties of high-resolution images. To address this gap, this paper proposes a novel hybrid method, termed mVDSR-GMM, which leverages the complementary strengths of Gaussian Mixture Models (GMM) and a modified Very Deep Super-Resolution (mVDSR) network. GMM is utilized to model the underlying statistical and probabilistic properties of the image for accurate high-frequency reconstruction while the mVDSR network enhanced with 25 convolutional layers and Leaky ReLU activations boosts the model’s structural feature extraction and non-linear mapping capabilities. This hybrid approach technically integrates the mVDSR and GMM through a parallel merging of their estimated residuals within the luminance channel, thereby enhancing the spatial reconstruction of satellite imagery. The enhanced mVDSR network, with increased depth and Leaky ReLU activation functions, further boosts the model's learning capability and robustness. Through extensive evaluations using satellite imagery datasets and metrics such as PSNR, SSIM, BRISQUE, MSE, UIQI, SAM, and FD, mVDSR-GMM consistently outperformed existing techniques, demonstrating its effectiveness in improving spatial resolution. Experimental results indicate that mVDSR-GMM outperforms main competitors such as SwinIR and SinSR across all evaluation metrics, providing a robust reconstruction performance that is further validated by improved accuracy in the binary separation of urban structures (buildings and roads) from natural environments. The proposed mVDSR-GMM emerges as a promising tool for advancing satellite imagery resolution, with potential applications in remote sensing and geospatial analysis.

  İncele
• MakaleAçık Erişim2026/r/gumushane/13603
  A Deep Learning–Based Framework for Automatic Determination of Developmental Dysplasia of the Hip from Graf Angles

  Taşolar, Sevgi; Günen, Mehmet Akif; Sığırcı, Ahmet; Taşolar, Hakan

  Developmental dysplasia of the hip (DDH) is a common neonatal condition that necessitates early diagnosis to ensure effective treatment. The traditional Graf method, while widely used for evaluating infant hips via ultrasound, is limited by operator dependency and measurement variability. This research has proposed a framework using deep learning network, morphological operation and local maxima method to diagnose DDH in newborns using ultrasound images. The method utilizes DeepLabv3 + for image segmentation, evaluating multiple backbone architectures (ResNet50, InceptionResNetV2, MobilenetV2, and Xception) to identify the region of interest accurately. Local maxima method was used to determine the extremum points of the line defining the Graf angles. Denoising filters, including mean, median, and Wiener, are applied to determine local maxima points accurately. The evaluation comprises two stages: first, assessing the performance of DeepLabv3 + backbones in producing masks for Graf angles determination, and second, comparing the angles obtained through proposed framework with those determined by expert radiologists. Comparative analysis demonstrates that MobileNetV2 (94.64 accuracy, 86.99 Cohen’s kappa, 94.31 F-score) surpasses other models in segmentation accuracy and measurement reliability. This conclusion is backed by key performance metrics such as accuracy, IoU, PSNR, F-score, SSIM, Cohen’s kappa, as well as by the intraclass correlation coefficient and Bland–Altman analyses. The proposed framework shows considerable promise in automating hip ultrasound analysis for DDH diagnosis, minimizing operator dependency while enhancing measurement consistency.

  • 
  İncele
• MakaleYalnız künye2026/r/gumushane/13554
  Comparison 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.

  • 
  İncele
• MakaleAçık Erişim2026/r/gumushane/12234
  Assessing 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.

  • 
  İncele
• MakaleYayıncıda açık erişim2025/r/gumushane/14300
  Optimizing Visibility of Historical Structures Using Modified Weighted Differential Evolution: Insights from the Kromni Valley, Gümüşhane, Türkiye

  Günen, Mehmet Akif; Öztürk, Kaşif Furkan; Aliyazıcıoğlu, Şener

  It is very important for historical structures to see each other in order to reveal the historical and cultural identity of a region. Historical structures in the Kromni Valley of Gümüşhane, located near the Sümela Monastery, served as places of worship, communication, trade, and social activity centers during their period of active use. This study analyses the spatial relationships of 38 historic buildings, including churches, chapels and castles, whose 3D models are created by in-situ measurements and point clouds obtained by unmanned aerial vehicles, using a 3D viewshed analysis using geographic information systems and remote sensing data. The research introduces a modified weighted differential evolution-based viewshed analysis (mWDE-WS) to enhance the visibility of these structures. In order to assess the applicability of the proposed method, a statistical comparison was conducted between four different Differential Evolution (DE) algorithms (standard DE, LSHADE, CobiDE, JADE and WDE) and the mWDE. The Wilcoxon signed-rank test indicates that mWDE is a more effective solution than alternative methods for addressing the relevant real-world issues. The study also integrates drainage network analysis to assess flood risks and the relationship between cultural structures and water flow. Findings show that historical structures in the region were built not randomly but within a rational approach and 64% of the study area is visible from structures and 2% of the area is visible from ten or more structures. mWDE-WS analysis revealed that the visible area could increase by 20% to 84.37% if the historic structures were placed in optimal locations. In addition, the historical structures were built away from 3rd order streams to minimize flood risk and humidity, demonstrating the community's awareness of the local topography and hydrology.

  • 
  İncele
• MakaleYalnız künye2025/r/gumushane/14283
  Discontinuities identification from rock outcrop using auto-encoder and point clouds

  Günen, Mehmet Akif; Aliyazıcıoğlu, Şener

  Rock slopes and geotechnical structures often exhibit discontinuity planes, which significantly influence their mechanical behavior and stability. The precise identification of these discontinuities is critical for ensuring the safety and cost-effectiveness of construction projects. However, previous methods are often limited by subjectivity, low automation, and challenges in inaccessible or hazardous environments. This study introduces a novel seven-step approach utilizing auto-encoders for estimating the orientation parameters of individual discontinuity sets from point cloud data. The steps include: data collection, filtering point clouds to eliminate outliers, determining neighborhood size, feature extraction, training a stacked auto-encoder for feature learning, estimating discontinuity sets using density-based clustering, and calculating individual discontinuities through the least squares method for orientation parameter estimation. In addition, the methodology incorporates a planarity-based approach for estimating neighborhood size. The statistical performance has been validated through evaluations in two real case studies and one synthetic case study. The classification achieves an accuracy exceeding 95%, highlighting the high efficacy of the approach. Results for orientation parameter estimation show consistency with existing methods and in situ measurements, with average orientation values for discontinuities within 5 degrees for individual sets. This framework represents a considerable advancement in rock discontinuity analysis, offering a robust and automated solution that is particularly effective in hazardous and hard-to-reach areas.

  İncele