Sürdürülebilir Kalkınma Amaçları

• 3
• 3
• 2
• 1

Araştırmacılar(6)

• Alici-Karaca, Demet
• Aslay, Fulya
• Aydın, Yıldız
• Özdemir, Ahmet Ozan
• Uğurtürk, Nihat Mert
• Yildiz, Eyyup

• ArticleAçık Erişim2025/r/ebyu/19568
  Vision Transformer-Based Approach: A Novel Method for Object Recognition

  Aydın, Yıldız; Ali, Ali Khudhair Abbas

  This paper proposes a hybrid method to improve object recognition applications on inefficient and imbalanced datasets. The proposed method aims to enhance object recognition performance using the Vision Transformer (ViT) deep learning model and various classical machine learning classifiers (LightGBM, AdaBoost, ExtraTrees, and Logistic Regression). The Caltech-101 dataset used in the study is a low-resolution and noisy image dataset with class imbalance problems. Our method achieves better results by combining the feature extraction capabilities of the Vision Transformer model and the robust classification performance of classical machine learning classifiers. Experiments conducted on the Caltech-101 dataset demonstrate that the proposed method achieves a precision of 92.3%, a recall of 89.7%, and an accuracy of 95.5%, highlighting its effectiveness in addressing the challenges of object recognition in imbalanced datasets.

  İncele
• ArticleAçık Erişim2025/r/ebyu/19550
  Enhanced emotion recognition through hybrid deep learning and svm integration

  Aydın, Yıldız; Türkeş, Muhammed Kerem

  The facial expression recognition system, which contributes to the processes to be more effective and faster in many fields such as medicine, education and security, plays an important role in various applications. For example, while emotional and psychological states can be monitored thanks to facial expression recognition in the health field, it can be used in critical applications such as lie detection in the security sector. In education, students' instant facial expressions are analyzed to contribute to the learning processes. The problem of emotion recognition from facial expressions, which is related to many fields, is of great importance in obtaining accurate and reliable results. Therefore, in order to increase the performance of emotion recognition from facial expressions, a hybrid approach combining deep learning and classical machine learning methods is considered in this study. In the proposed method, the ResNet50 model is used as a feature and Support Vector Machines (SVM) is used as a classifier. In this study, a hybrid approach consisting of the combination of ResNet50 and SVM methods is proposed-to increase the performance of emotion recognition from facial expressions. In order to analyze facial expressions, six basic emotions are classified as happiness, sadness, anger, fear, surprise and disgust using the CK+48 dataset. Experimental results show that the proposed hybrid approach has high accuracy in emotion recognition and outperforms traditional machine-learning algorithms.

  İncele
• ArticleAçık Erişim2025/r/ebyu/19548
  Comparison of the effects of features and classifiers on performance in the cardiovascular disease detection system

  Emir, İzzet; Aydın, Yıldız

  This study aims to analyze the effects of features and classifiers in detecting cardiovascular diseases (CVD), which remain the leading cause of morbidity and mortality worldwide. Early and accurate detection of CVD significantly affects treatment outcomes. Therefore, the proposed method aims to automatically detect cardiovascular diseases via artificial intelligence. In this research, the performances of artificial intelligence methods for the cardiovascular disease detection problem are analyzed. The dataset used in this study was sourced from the publicly available Kaggle platform. It used for performance analysis in the developed application includes the features of 70000 patients such as age, gender, height, weight, blood pressure, cholesterol, glucose, smoking and alcohol use. These features were classified with Gradient Boosting, XGBoost, SVM, Random Forest, Logistic Regression, Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM) methods and performance comparison was performed. In the experimental results, the highest accuracy rate of 72.55% was obtained using the Gradient Boosting method, demonstrating its superior performance in cardiovascular disease detection. In addition, it was observed that the classification performance decreased when the high blood pressure attribute was removed from the dataset, while the removal of other features did not significantly affect the performance.

  • 
  İncele
• ArticleAçık Erişim2024/r/ebyu/19880
  An unconditional generative model with self-attention module for single image generation

  Yuksel, Mehmet Erkan; Sevgen, Selcuk; Yıldız, Eyyüp

  Generative Adversarial Networks (GANs) have revolutionized the field of deep learning by enabling the production of high-quality synthetic data. However, the effectiveness of GANs largely depends on the size and quality of training data. In many real-world applications, collecting large amounts of high-quality training data is time-consuming, and expensive. Accordingly, in recent years, GAN models that use limited data have begun to be developed. In this study, we propose a GAN model that can learn from a single training image. Our model is based on the principle of multiple GANs operating sequentially at different scales, where each GAN learns the features of the training image and transfers them to the next GAN, ultimately generating examples with different realistic structures at the final scale. In our model, we utilized a self-attention and new scaling method to increase the realism and quality of the generated images. The experimental results show that our model performs image generation successfully. In addition, we demonstrated the robustness of our model by testing it in different image manipulation applications. As a result, our model can successfully produce realistic, high-quality, diverse images from a single training image, providing short training time and good training stability.

  İncele
• ArticleAçık Erişim2024/r/ebyu/19817
  Geliştirilmiş EfficientNet-B0 mimarisi ile Helikobakter Pilorinin Teşhisi

  Nalbantoglu, Ozkan Ufuk; Basturk, Alper; Akay, Bahriye Basturk; Alici Karaca, Demet; Karaboga, Dervis

  Kansere bağlı ölümlerde önde gelen türlerden olan mide kanserine çevresel ve genetik birçok faktör sebebiyet verebilir. Başlıca risk faktörlerinden birisi ise midede gastrit ve ülsere neden olan helikobakter pilori bakteri virüsüdür. Bu virüsün tespit edilebilmesi için histopatolojik değerlendirme yapılmaktadır. Manuel yapılan bu işlem iş yükü, zaman kaybı ve subjektif değerlendirmeden kaynaklı patologlar arası görüş ayrılıklarına sebebiyet vermektedir. Tanı sürecini hızlandırmak ve hastaya zamanında tedavi uygulayarak yaşam süresini uzatmak amacıyla otomatik sistemlere ihtiyaç duyulmaktadır. Bu çalışmada son yıllarda başarımı artarak devam eden derin öğrenme mimarisi histopatolojik tam slayt görüntüden helikobakter pilorinin varlığını teşhis etmek için kullanılmaktadır. Mide biyopsi görüntülerini içeren halka açık DeepHP veri seti kullanılarak Helikobakter pilorinin tanısında uçtan-uca bir derin öğrenme modeli olanEfficientNet-B0 uygulanmıştır. Ayrıca, ağın özellik çıkarma yeteneğini geliştirmek amacıyla son zamanlarda literatüre sunulan çeşitli dikkat mekanizmaları (Etkili Kanal Dikkat, Frekans Kanal Dikkati Ağı, Kapılı Kanal Dönüşümü, Evrişimsel Blok Dikkat Modülü ve Basit, Parametresiz Dikkat Modülü) derin modele entegre edilerek model başarımı üzerindeki etkileri incelenmiştir. Yapılan analizler sonucunda, Frekans Kanal Dikkat Ağı entegre edilen EfficientNet-B0 mimarisinin, histopatolojik görüntülerden helikobakter pilorinin tanısında 0.99835 doğruluğa ulaştığı görülmüştür. Buna göre, önerilen model literatürde yer alan modellerin DeepHP veri seti üzerinde ürettiği sonuçlardan çok daha üstün bir sonuç üretmiştir ve hastalığın tanısında umut vaat edicidir.

  • 
  İncele
• ArticleAçık Erişim2024/r/ebyu/19661
  Investigating the effect of loss functions on single-image GAN performance

  Yuksel, Mehmet Erkan; Sevgen, Selcuk; Yıldız, Eyyüp

  Loss functions are crucial in training generative adversarial networks (GANs) and shaping the resulting outputs. These functions, specifically designed for GANs, optimize generator and discriminator networks together but in opposite directions. GAN models, which typically handle large datasets, have been successful in the field of deep learning. However, exploring the factors that influence the success of GAN models developed for limited data problems is an important area of research. In this study, we conducted a comprehensive investigation into the loss functions commonly used in GAN literature, such as binary cross entropy (BCE), Wasserstein generative adversarial network (WGAN), least squares generative adversarial network (LSGAN), and hinge loss. Our research focused on examining the impact of these loss functions on improving output quality and ensuring training convergence in single-image GANs. Specifically, we evaluated the performance of a single-image GAN model, SinGAN, using these loss functions in terms of image quality and diversity. Our experimental results demonstrated that loss functions successfully produce high-quality, diverse images from a single training image. Additionally, we found that the WGAN-GP and LSGAN-GP loss functions are more effective for single-image GAN models.

  İncele
• ArticleAçık Erişim2024/r/ebyu/19645
  Block-Based Forgery Detection: Performance Comparison Using HOG, LBP, and MBF

  Babacan, Yunus; Aydın, Yıldız

  One of the types of forgery performed on digital images is copy and paste forgery (CPS). This type of forgery is realized by pasting another region copied from the same image over the relevant region of the image. It is very important to determine whether there is any forgery on these images, which can be used as evidence in many fields. In this study, an analysis on forgery detection is performed using HOG (Histogram of Oriented Gradients), LBP (Local Binary Patterns), and Multiscale Basic Features (MBF) features for block-based copy-paste forgery detection. The performance of various features alone and in combination is evaluated. Combinations such as HOG+LBP, HOG+MBF and MBF+LBP were tried, but the expected performance improvement was not achieved. Although the performance increase is not very high, the highest results are generally obtained with the LBP+MBF hybrid feature This approach resulted in an F1 score of 88.5%. This study contributes to existing methods in the field of block-based forgery detection and demonstrates the effectiveness of various feature combinations. In addition, although HOG and LBP features are frequently used in block-based approaches, approaches using the MBF feature have not been found in the literature. This study contributes to the existing methods in the field of block-based forgery detection and shows the effectiveness of various features and feature combinations.

  İncele
• ArticleAçık Erişim2024/r/ebyu/19638
  A Mobile Secure Content Development Application for Children in the Software Development Life Cycle Framework

  Aslay, Fulya; Karabaş, İsakhan; Öksüz, Elif

  This study deals with the software development process of a mobile application that will support children's learning, development and entertainment processes while ensuring their safety in the online environment. The project was carried out within the framework of the Software Development Life Cycle (SDLC) and adopted a systematic approach including requirements analysis, design, development, testing and maintenance phases. The application aims to provide appropriate content for children by minimizing digital security risks and to enable parents to safely control their children's internet use. With the Extreme Programming (XP) methodology, a continuous improvement process based on user feedback was implemented and security and performance tests were successfully completed. The article shows that not only coding but also effective management of the whole process in software development projects contributes to project success. The study is expected to contribute to the literature on children's digital safety and provide a basis for future research in this field.

  • 
  İncele
• ArticleAçık Erişim2023/r/ebyu/20216
  Sentiment Analyzing from Tweet Data’s Using Bag of Words and Word2Vec

  Aydın, Yıldız

  Twitter sentiment classification is an artificial approach for examining textual information and figuring out what people's publicly tweets from a variety of industries are experiencing or thinking. For instance, a large number of tweets containing hashtags are posted online every minute from one user to some other user in the commercial and politics fields. It can be challenging for scientists to correctly comprehend the context in which specific tweet terms are used, necessitating a challenge in determining what is actually a positive or negative comment from the vast database of twitter data. The system's authenticity is violated by this issue and user dependability may be significantly diminished. In this study, twitter data sent to interpret movies were classified using various classifier and feature methods. In this context, the IMDB database consisting of 50000 movie reviews was used. For the purpose of anticipating the sentimental tweets for categorization, a huge proportion of twitter data is analyzed. In the proposed method, bag of words and word2vec methods are given by combining them instead of giving them separately to the classifier. With both the suggested technique, the system's effectiveness is increased and the data that are empirically obtained from the real world situation may be distinguished well. With experimental efficiency of 90%, the suggested approach algorithms' output attempts to assess the reviews tweets as well as be able to recognize movie reviews.

  İncele
• ArticleAçık Erişim2023/r/ebyu/20084
  Üniversite Öğrencilerinin Covid-19’a Yakalanma Riski ve Pandeminin Öğrenciler Üzerindeki Etkisinin Makine Öğrenmesi Yöntemleriyle Tahmin Edilmesi

  Aslay, Fulya; Önder, Şeyma; Demir, Fatma

  Bu çalışmada Erzincan Binali Yıldırım Üniversitesi öğrencileri üzerinde bir araştırma yapılmıştır. Makine öğrenmesi yöntemleri kullanılarak öğrencilerin Covid-19’a yakalanma risklerini tahmin eden bir model geliştirilmiştir ayrıca Covid-19’un öğrenciler üzerindeki etkisi araştırılmıştır. Bu amaçla 2022 yılı Mart ayında Erzincan Binali Yıldırım Üniversitesi öğrencilerine bir dijital anket uygulanmıştır. Anket verilerine göre öğrencilerin makine öğrenmesi yöntemlerinden Karar Ağaçları, K-En Yakın Komşu, Destek Vektör Makineleri ve Naive Bayes ile Covid-19’a yakalanma riskleri tahmin edilmiştir ve pandeminin öğrenciler üzerindeki etkileri araştırılmıştır. Tüm tahmin modelleri analizleri karşılaştırmalı olarak sunulmuştur. Çalışmada oluşturulan tahmin modellerinin başarısını ölçmek için ise R2, RMSE ve MAE metrikleri kullanılarak modellerin belirlilik katsayısı ve hata değerleri sunulmuştur. Buna göre destek vektör makineleri ile yapılan tahmin sonucunda R2 değeri 0,9323 ile öğrencilerin Covid-19’a yakalanma riskleri en başarılı şekilde tahmin edilmiştir.

  • 
  • 
  İncele