Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Ozer Ozdemir has a total of 15 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    HybridTopo: Topology-Aware Supervision and Inference for Connectivity-Preserving Road Network Extraction From Aerial Imagery

    by Ozer Ozdemir, Asli Kaya Karakutuk

    In this study, we address road network extraction from high-resolution satellite imagery with explicit emphasis on structural connectivity in addition to pixel-wise segmentation accuracy. Although modern deep learning models can achieve strong IoU values, topological discontinuities may still degrade routing usability. To address this issue, we present HybridTopo, a topology-aware supervision and inference framework that integrates topology-sensitive loss design, auxiliary centerline learning, and inference-time refinement for road extraction. Within the Massachusetts Roads benchmark used in this study, our analysis indicates that explicit topology-aware supervision is more strongly associated with structural preservation than the architectural variations evaluated here. All quantitative results are reported as mean ± standard deviation over three independent runs unless otherwise stated. Under the full pipeline, network fragmentation is reduced from 37.4 to 10.3 connected components per tile, corresponding to a 73% relative reduction with respect to the baseline configuration. In addition, APLS-lite evaluation based on random sampling of 64 node pairs indicates improved path-length consistency in the predicted road graphs. These findings highlight the trade-off between pixel accuracy and topological discipline and show that hybrid structural supervision can improve functionally meaningful road extraction on this benchmark, while broader cross-dataset generalization remains future work.

  • Open Access

    Optuna-Optimized Pythagorean Fuzzy Deep Neural Network: A Novel Framework for Uncertainty-Aware Image Classification

    by Asli Kaya Karakutuk, Ozer Ozdemir, Sevil Senturk

    By using Geographic Information Systems, satellite imagery from remote sensing techniques provides quantitative and qualitative data about Earth’s natural and human elements. However, the direct use of raw imagery may prevent the accurate identification of the spectral and temporal characteristics of the target objects. To obtain meaningful results from these data, the object and surface features in the image must be classified correctly. In this context, this study develops a new deep learning approach that includes hyperparameter optimization that considers uncertainty factors when classifying satellite imagery. In the proposed approach, a hybrid architecture called CNN-Pythagorean Fuzzy Deep Neural Network (PFDNN) is developed by combining the ability of convolutional neural networks (CNN) to reveal expressive features with the ability of Pythagorean fuzzy set (PFS) theory to predict uncertainty. In addition, to further improve the model’s success, the hyperparameters are automatically optimized using Optuna. In the experiments conducted on the EuroSAT RGB dataset, the CNN+PFDNN+Optuna model achieved 0.9696 ± 0.0037 accuracy and a macro-AUC value of 0.9983, outperforming other methods such as DNN, FDNN, PFDNN and VGG16+PFDNN. Including the Pythagorean fuzzy layer in the system provided about 13.05% higher accuracy than conventional fuzzy systems. These results show that integrating the Pythagorean fuzzy set approach into deep learning models contributes to more effective management of uncertainties in remote sensing data and that hyperparameter optimization significantly impacts model performance.


  • Open Access

    Veri Madenciliğinde Karar Ağaçları İndükleyicilerinin Likert Ölçekli Verilerde Sınıflandırma Performansı

    by Ozer Ozdemir, Ferdi Karakütük, Aslı Kaya Karakütük

    Madencilik terimi ile benzer anlam taşıyan veri madenciliği, sorunların çözülmesine, eğilimlerin tahmin edilmesine, risklerin azaltılmasına ve yeni fırsatlar bulunmasına yardımcı olmak için muazzam miktarda bilgi ve veri setini analiz etme, yararlı zekayı keşfetme sürecidir. Bu çalışmada veri madenciliğinin muazzam yeteneklerinden faydalanarak Likert ölçekli veri tiplerinde bilgi keşfi yapılması amaçlanmıştır. Farklı veri madenciliği tekniklerinin Likert ölçekli veri türleri üzerinde sınıflandırma başarısını karşılaştırmak üzere veri seti olarak Türkiye Aile Yapısı Araştırması (TAYA) seçilmiştir. Yapısı gereği dengesiz olan veri seti üzerinde ilk olarak dengesizlik giderilmeden sınıflandırma yapılmış ardından sınıflar arası dengesizlik giderilmiş ve sınıflama analizine etkisi gözlemlenmiştir. Sınıflar arası dengesizliği giderebilmek amacıyla yeniden örnekleme ve veri tamamlama yöntemi ile toplam örnek hacmi değiştirilerek üç farklı veri seti oluşturulmuştur. Oluşturulan veri setlerinde sınıflandırma başarısı en yüksek olan algoritmanın CART algoritması olduğu görülmüştür. Dengesizlik giderilmeden yapılan sınıflandırmada ise RepTree algoritmasının daha başarılı sonuçlar ürettiği görülmüştür.


  • Open Access

    A Novel Fuzzy Kernel Extreme Learning Machine Algorithm in Classification Problems

    by Asli Kaya Karakutuk, Ozer Ozdemir

    Today, numerous methods have been developed to address various problems, each with its own advantages and limitations. To overcome these limitations, hybrid structures that integrate multiple techniques have emerged as effective computational methods, offering superior performance and efficiency compared to single-method solutions. In this paper, we introduce a basic method that combines the strengths of fuzzy logic, wavelet theory, and kernel-based extreme learning machines to efficiently classify facial expressions. We call this method the Fuzzy Wavelet Mexican Hat Kernel Extreme Learning Machine. To evaluate the classification performance of this mathematically defined hybrid method, we apply it to both an original dataset and the JAFFE dataset. The method is enhanced with various feature extraction methods. On the JAFFE dataset, the algorithm achieved an average classification accuracy of 94.55% when supported with local binary patterns and 94.27% with a histogram of oriented gradients. Moreover, these results outperform those of previous studies conducted on the same dataset. On the original dataset, the proposed method was compared with an extreme learning machine and wavelet neural network, and it was found that the method has remarkable efficiency compared to the other two methods.


  • Comparison of five scores to predict mortality in malignant pleural effusion

    by Merve Ayik Türk, Gülru Polat, Özer Özdemir, Yunus Türk, Berna Kömürcüoğlu

    Malignant pleural effusion (MPE) is a complication of malignancy. Treatment of MPE is based on predicted outcome. The aim of this study was to compare the performance characteristics of LENT, PROMISE, RECLS, AL and pNLR scores for prediction of mortality in lung cancer patients who have MPE. Patients who were diagnosed with MPE that was associated with underlying lung cancer between January 2010 and December 2019 were included and analyzed retrospectively in a single center. Outcomes considered were 30-day, 6 months, and 1-year mortality. A total of 180 patients were examined. For 30-day mortality, the areas under the ROC curves (AUC) (95% CI) were: LENT 0.83 (0.76–0.87), RECLS 0.71 (0.63–0.77), and PROMISE 0.70 (0.17–0.38). For 6-month and 1-year mortality the order of these AUCs was similar. Cox regression showed that none of the scores were significantly associated with 30-day mortality, but LENT and RECLS were significantly associated with 6-month and 1-year mortality. Comparison of – 2log likelihood ratios showed that LENT score was more, strongly associated with 6-month mortality than PROMISE (p = 0.001) or RECLS (p = 0.02). LENT score was also more strongly associated with 1-year mortality than PROMISE (p = 0.001) but there was no difference between LENT and RECLS score (p = 0.64). We observed that the LENT score was more predictive than the other scores in mortality in patients who have lung cancer and MPE. The LENT and RECLS scores have similar performance characteristics for prediction of 1-year mortality in these patients.


  • Open Access

    Sociodemographic and clinical risk factors associated with in-hospital tuberculosis mortality in Türkiye, 2008-2018

    by Mine Gayaf, Merve Ayik Türk, Özer Özdemi̇r, Gülru Polat, Onur Karaman, Filiz Güldaval, Gülsüm Ari, Dursun Tatar, Ahmet Emin Erbaycu

    Introduction: Tuberculosis (TB) is an infectious disease that can be fatal if left untreated or poorly treated, and it is associated with many morbidities. Deaths may provide better understanding of the associated factors and help guide interventions to reduce mortality. In this study, it was aimed to reveal some of the features that predict hospital mortality in patients with TB and to present some alarming findings for clinicians. Materials and Methods: Patients who had been hospitalized with the diagnosis of TB between January 2008 and December 2018 were included and analyzed retrospectively. In -hospital mortality because of any TB disease after the initiation of treatment in patients admitted to the TB Ward and the primary cause of mortality were taken as endpoint. Results: A total of 1321 patients with a mean age of 50.1 years were examined. Total mortality was 39.4% (521 deaths) and 13.1% were in -hospital deaths (173 deaths). Of the deaths, 61.8% (n= 107) occurred during the first month after TB treatment were started. On univariate analysis, age over 48.5 years, Charlson comorbidity index, extension of radiological involvement, hypoalbuminemia and lymphopenia were most predictive variables with higher odds ratios (respectively, p< 0.001 for all). Conclusion: In -hospital tuberculosis disease mortality is related with older age, cavitary or extensive pulmonary involvement, low albumin levels, unemployment, cigarette smoking and especially those with concomitant malignancy and chronic pulmonary disease.


  • Open Access

    İstanbul ve Eskişehir Bölgesi Minimum-Maksimum Yağış Miktarlarının Ekstrem Değerler Dağılımları Ailesi ile Modellenmesi

    by Mustafa Çavuş, Özer Özdemir, Ahmet Sezer

    Ekstrem (uç) olaylar doğada az olasılıkla ortaya çıkmakla birlikte, ortaya çıktıklarında etki alanları hem ekonomik hem de çevresel olarak büyük olmaktadır. Bu yüzden bu tür olayların hangi olasılıkla ortaya çıkabileceği, bu konuda alınması gereken tedbirlerin neler olması gerektiği konusunda belirleyici olmaktadır. Ekstrem(uç) değerler dağılımları ailesindeki dağılımların parametrelerinin doğru belirlenmesi ekstrem olasılıkların doğru hesaplanmasına katkıda bulunacaktır. Literatürdeki Weibull, Frechet, Gumbell dağılımlarına ilişkin parametreler R programında yer alan fitdistrplus fonksiyonu kullanılarak elde edilecektir. Bu çalışmada ülkemiz nüfusunun yoğun olarak bulunduğu İstanbul bölgesi ve Eskişehir bölgesindeki maksimum ve minimum yağış miktarlarının dağılımı blok maksimum yöntemi kullanılarak araştırılacaktır. Bu dağılımlardan elde edilecek sonuçlar, potansiyel risklere ve fırsatlara karar verilmesine yardımcı olacaktır.