TR-Dizin İndeksli Yayınlar Koleksiyonu
Permanent URI for this collectionhttps://hdl.handle.net/20.500.12573/396
Browse
17 results
Search Results
Article Makine Öğrenmesi ve Derin Öğrenme Yöntemleri ile Hidroponik Tarım(2023) Bulut, Nurten; Hacıbeyoglu, MehmetIn the face of the rapidly increasing population of our world today, researchers have turned to studies that use existing resources more effectively and efficiently in addition to searching for new resources in order to meet the rapidly decreasing needs such as raw materials and nutrients. The use of hydroponic agriculture, which is one of the alternative methods that can be used to meet the need for nutrients, which is one of the greatest needs of humanity, has become more popular day by day. The use of nutrient solution water instead of soil, the fact that it is not affected by weather conditions, that it can be applied indoors and that it can be vertically oriented are the characteristics that make hydroponic agriculture different from other agricultural methods. In addition, the lack of soil in this agricultural method brings with it the need for more observation and supervision. The aim of this study is to show that the observation and surveillance needs necessary to increase yield in hydroponic agriculture can be achieved using machine learning and deep learning methods. For this purpose, it has been observed that the efficiency of hydroponic agriculture has been increased in experimental studies conducted using five machine learning and deep learning methods. The deep learning method has achieved better results with 99.7% success compared to other methods.Article Karanlık Videolarda Orman Yangını Tespitine Yönelik Uzaysal-Zamansal Nesne Öznitelikleri(2022) Agirman, Ahmet Kerim; Tasdemir, KasimIn this paper, a wildfire detection algorithm from dark videos is proposed. Unlike the daytime wildfires, in the dark videos, neither the fire nor its surrounding has visually clearly perceptible texture. Its unique visual characteristics make it challenging to extract descriptive object features. This paper addresses the challenging problem by tracking the glowing objects in the darkness and extracting features based on the spatio-temporal behavior of them. It is experimentally shown that the proposed features are descriptive enough to classify wildfires with over 90% accuracy even there exists deceptive light sources such as city lights, flashlights, car headlights and reflections in the scene. Moreover, we investigate several conventional machine learning algorithms such as ensemble and kernel-based methods on the same spatio-temporal feature set. Comprehensive empirical test results demonstrate that the most accurate detection is obtained when the spatio-temporal feature set is classified using Random Forest.Article İşbirlikçi Filtreleme temelinde Film Öneri Sistemleri: Netflix üzerinde bir VakaÇalışması(2021) Sütçü, Muhammed; Kaya, Ecem; Erdem, OğuzkanFilmler, şarkılar ve alışveriş ürünleri gibi ögelerin kullanıcı değerlendirmeleriÖneri Sistemleri (ÖS) tarafından henüz değerlendirilmemiş ürünleri tahmin etmekiçin kullanılır. ÖS kullanıcılara çeşitli alanlarda öneri vermek için geliştirilmiştir veÖS uygulama alanlarından birisi de film önerisidir. Bu alanda üç genel algoritmakullanılmaktadır; kullanıcılar arası benzerliğe dayanarak tavsiye veren İşbirlikçiFiltreleme, kullanıcı-eşya eşleştirilmesindeki ilişkiden beslenen İçerik TabanlıFiltreleme ve bu iki algoritmayı birleştiren Hibrit Filtreleme. Bu çalışmamızdaİşbirlikçi Filtreleme çerçevesinde hangi metotların daha etkili çalıştığı incelenmiştir.Analizimizde Netflix Ödül veri seti kullanılmış ve iyi bilinen İşbirlikçi Filtrelememetotları olan Tekil Değer Ayrışımı, Tekil Değer Ayrışımı++, K En Yakın Komşu veEş Kümeleme kıyaslanmıştır. Her metodun hatası Ortalama Hata Kare Kökükullanılarak ölçülmüştür. Son olarak, K En Yakın Komşu metodunun veri setimizdedaha başarılı olduğu sonuçlanmıştır.Article Benchmarking Deep Learning Models for Dermatological Image Analysis: EfficientNet Takes the Lead(2024) Kaçmaz, Rukiye Nur; Doğan, Refika SultanSkin cancer that spreads quickly and is deadly is called melanoma. If skin cancer is not treated in its early stages, the mortality rate is very high, but when it is correctly identified in its early stages, patients' lives can be saved. With an accurate and fast diagnosis, the patient's chance of survival can be increased. A computer- aided diagnostic support system needs to be created. In this study, Dense201, DarkNet19, and EfficientNet offer 3 different deep transfer learning models for melanoma classification. In addition, an ablation study was conducted in terms of the filter size used in transfer learning. To look at the effect of the filter size, different filter sizes were created in each model and the results were obtained. The ISIC dataset containing 1792 benign and 1464 malignant images was used in the study. According to this study, DenseNet201 provided accurate and reliable results at different filter sizes regardless of their size. Therefore, it is recommended to use DenseNet201 in studies involving the classification of skin lesions.Article LSTM Mimarisiyle Bilgilendirici Göğüs Röntgeni Altyazıları Oluşturma(2025-07-31) Bakal, Mehmet; Güzel, Ömer Faruk; Tanrıverdi, HarunBiyomedikal görüntüleme, tıp uzmanları için en etkili tıbbi tarama prosedürüdür. Özellikle röntgen görüntüleri, tıbbi teşhis amacıyla yoğun bir şekilde referans noktası olarak kullanılmaktadır. Ancak, röntgen görüntülerinden altta yatan hususları anlamak önemli radyolojik bilgi gerektirir. Bu çalışmada, bir kodlayıcı modülü olarak DenseNet121 sinir ağı mimarisini ve kelime gömme katmanları olarak metinsel veri (altyazılar) öğelerini kullanan bir derin öğrenme modeli, verilen X-ışını görüntülerinin ilgili başlık / altyazı bilgilerini tahmin etmek için eğitilmiştir. Oluşturulan model, özellikle nöral makine çevirisi görevleri için kullanılan tipik bir diziden diziye modeldir. Deneylerde, eğitim ve test aşamaları için Indiana Üniversitesi tarafından hazırlanan Open-i veri tabanı kullanılmıştır. Veri kümesi, bir alan uzmanı tarafından oluşturulan XML formatında saklanan 7.470 X-ray görüntüsü ve 3.955 hasta raporundan oluşmaktadır. Metinsel raporlar izlenimler, bulgular, karşılaştırmalar ve endikasyonlar dahil olmak üzere dört özel başlık içermektedir. Model geliştirme sırasında, izlenim başlıkları altındaki metinsel verilerden eğitim ve test adımlarında yararlanılmıştır. Modelin performansını ölçmek için İki Dilli Değerlendirme Alt Çalışma Puanı (BLUE) hesaplanmış ve birincil performans değerlendirme metriği olarak kullanılmıştır. BLUE puanlarına göre en iyi performans puanı, diğer n-gram setlerine kıyasla (burada n: 1, 2 ve 3) 0,38368 BLUE puanı ile dört kelime (dört gram) tahmin edildiğinde elde edilmiştir. Bu araştırma çalışması, otomatik teşhis amaçlı tıbbi görüntü veri kümelerinde metin oluşturma görevinde diziden diziye modellerin gücünü göstermektedir.Article Forecasting the Consumer Price Index in Türkiye Using Machine Learning Models: A Comparative Analysis(Gazi Univ, 2025-09-01) Söylemez, İsmet; Ünlü, Ramazan; Nalici, Mehmet ErenThis study utilizes machine learning models to forecast Türkiye's Consumer Price Index (CPI), thereby addressing a critical gap in inflation prediction methodologies. The central research problem involves the forecasting of CPI in a volatile economic environment, which is essential for informed policymaking. The primary objective of this study is to evaluate the performance of three machine learning models, such as Decision Tree (DT), Random Forest (RF), and Support Vector Machine (SVM), in forecasting CPI over periods ranging from one to six months, utilizing data from 2012 to 2024. The study's unique contribution lies in the application of the \"SelectKBest\" method, which identifies the most relevant indices, thereby enhancing the efficiency of the models. An ensemble method, Averaging Voting, is also employed to combine the strengths of these models, producing more accurate and robust predictions. The findings indicate that while the RF model consistently generates the most accurate forecasts across all shifts, the SVM model demonstrates a particular strength in the domain of short-term predictions. The ensemble model demonstrates a substantial performance improvement, with a R2 value of 0.962 for one-month ahead of estimates and 0.956 for five-month forecasts. This combined approach has been shown to outperform individual models, offering a more reliable framework for CPI forecasting. The findings offer valuable insights for economic policymakers, enabling more precise and stable inflation predictions in Türkiye.Article Optimizing Parameters for Efficient Computation With Fully Homomorphic Encryption Schemes(Tubitak Scientific & Technological Research Council Turkey, 2025-03-21) Karaagac, Cavidan Yakupoglu; Rohloff, Kurt; Yakupoğlu Karaağaç, Cavidan; Yakupoglu, CavidanIn this study, we aim to provide a parameter selection approach for the BFVrns scheme, one of the prominent fully homomorphic encryption (FHE) schemes. Selecting parameters for lattice-based FHE schemes poses a practical challenge for both experts and nonexperts. To solve this problem, we introduce a hybrid approach that combines theoretical approach with experimental analysis. First, we employ regression analysis to examine the impact of parameters on both performance and security. The varying behavior of FHE parameters in terms of performance, security, and ciphertext expansion factor (CEF) makes parameter selection more challenging. To address this issue, we employ a multi-objective optimization algorithm to determine the optimal parameter set for performance, CEF, and security simultaneously. As a result of this optimization, we obtain an improved parameter set that enhances performance at a given security level while ensuring correctness and resistance to lattice-based attacks, maintaining at least 128-bit security. Our results achieve an average similar to 5x reduction in CEF and generally better performance compared to the parameter sets in a previous BFVrns study. Our approach serves as a semi-automated parameter selection method for the PALISADE homomorphic encryption library, a widely recognized FHE library. This study sets a precedent for other FHE libraries.Article Machine Learning Based Early Prediction of Type 2 Diabetes: A New Hybrid Feature Selection Approach Using Correlation Matrix With Heatmap and SFS(2022-04-30) Buyrukoglu, Selim; Akbaş, AyhanA new hybrid machine learning method for the prediction of type 2 diabetes is introduced and explained in detail. Also, outcomes are compared with similar researches. Early prediction of diabetes is crucial to take necessary measures (i.e. changing eating habits, patient weight control etc.), to defer the emergence of diabetes and to reduce the death rate to some extent and ease medical care professionals’ decision-making in preventing and managing diabetes mellitus. The purpose of this study is the creation of a new hybrid feature selection approach combination of Correlation Matrix with Heatmap and Sequential forward selection (SFS) to reveal the most effective features in the detection of diabetes. A diabetes data set with 520 instances and seven features were studied with the application of the proposed hybrid feature selection approach. The evaluation of the selected optimal features was measured by applying Support Vector Machines(SVM), Random Forest(RF), and Artificial Neural Networks(ANN) classifiers. Five evaluation metrics, namely, Accuracy, F-measure, Precision, Recall, and AUC showed the best performance with ANN (99.1%), F-measure (99.1%), Precision (99.3%), Recall (99.1%), and AUC (99.2%). Our proposed hybrid feature selection model provided a more promising performance with ANN compared to other machine learning algorithms.Article Citation - WoS: 2Machine Learning Based Network Intrusion Detection With Hybrid Frequent Item Set Mining(Gazi Univ, 2024-10-02) Firat, Murat; Bakal, Gokhan; Akbas, Ayhan; Bakal, MehmetWith the development and expansion of computer networks day by day and the diversity of software developed, the damage that possible attacks can cause is increasing beyond the predictions. Intrusion Detection Systems (STS/IDS) are one of the practical defense tools against these potential attacks that are constantly growing and diversifying. Thus, one of the emerging methods among researchers is to train these systems with various artificial intelligence methods to detect subsequent attacks in real time and take the necessary precautions. However, the ultimate goal is to propose a hybrid feature selection approach to improve the classification performance. The raw dataset originally enclosed 85 descriptor features (attributes) for classification. These attributes are extracted using CICFlowMeter from a PCAP file where network traffic is recorded for data curation. In this study, classical feature selection methods and frequent item set mining approaches were employed in feature selection for constructing a hybrid model. We aimed to examine the effect of the proposed hybrid feature selection approach on the classification task for the network traffic data containing ordinary and attack records. The outcomes demonstrate that the proposed method gained nearly 3% improvement when applied with the Logistic Regression algorithm on classifying more than 225,000 records.Article Evaluation of Sub-Network Search Programs in Epilepsy-Related GWAS Dataset(Pamukkale Univ, 2022) Adanur Dedeturk, Beyhan; Bakir Gungor, Burcu; Dedeturk, Beyhan Adanur; Gungor, Burcu BakirThe active sub-network detection aims to find a group of interconnected genes of disease-related genes in a protein-protein interaction network. In recent years, several algorithms have been developed for this problem. In this study, the analysis of disease-specific sub-network identification programs is evaluated using epilepsy data set. Under the same conditions and with the same data set, 9 different programs are run and results of their Greedy algorithm, Genetic algorithm, Simulated Annealing Algorithm, MCC (Maximal Clique Centrality) algorithm, MCODE (Molecular Complex Detection) algorithm, and PEWCC (Protein Complex Detection using Weighted Clustering Coefficient) algorithm are shown. The top-scoring 5 modules of each program, are compared using fold enrichment analysis and normalized mutual information. Also, the identified subnetworks are functionally enriched using a hypergeometric test, and hence, disease-associated biological pathways are identified. In addition, running times and features of the programs are comparatively evaluated.