18 results
Search Results
Now showing 1 - 10 of 18
Conference Object Citation - Scopus: 4Convolution Neural Network (cnn) Based Automatic Sorting of Cherries(Institute of Electrical and Electronics Engineers Inc., 2021) Park,H.; Khan,M.U.Cherries are spring fruits enriched with nutrients, and are easily available in food markets around the world. Due to their excess demand, many enterprises solely focused on their processing. Cherries are especially susceptible to pathological-, physiological-diseases and structural degradation due to their soft outer skin. The post-harvest life of the fruit is limited by various characteristics. The agricultural industry has also been at the forefront to get benefits from the advanced machine learning tools. This study presents an image processing-based system for sorting cherries using the convolutional neural network (CNN). For this study, Prunus avium L cherries of export quality, available in Turkey, tagged as ‘0900 Ziraat’, are used. Surprisingly, there exists no dataset for these cherries; hence, we developed our dataset. Through the proposed approach based upon U-Net, the binary classification accuracy of 99% is achieved. Clear identification is demonstrated by the test results of varying mixture ratios of good and bad cherries. It can therefore be said that for cherry sorting and grading, U-Net can be applied as a reliable and promising machine learning tool. ©2021 IEEEMaster Thesis Makine Öğrenmesi Algoritmalarının Oyun Seviyelerinin Zorluklarının Belirlenmesinde Kullanılması(2020) Şahbenderoğlu, Turan Ozan; Karakaya, ZiyaIn game design, adjusting difficulty is one of the key aspects of financial success. However, this task is costly since it is time-consuming. In the literature, there are very limited studies according to determining the game difficulty. Instead, almost every study is about difficulty adjustment which skips the determining process. This thesis aims to develop a game environment to observe if the machine learning can determine the difficulty of a game and the game levels. For this purpose, a game with five different levels from easy to hard is developed in Unity Engine. A machine learning agent that uses reinforcement learning is also developed and each game level used as learning environment of the agent. In general, the learning process shows that the Cumulative Reward of the agents is decreased as levels become harder. The complexity of the game significantly decreases Cumulative Rewards. The results of this thesis have shown that those level difficulties of a game can be determined by comparing the reinforcement learning agent's performance on collecting rewards in the training area. In other words, machine learning algorithms have a big potential to support the game design phase of the game development process when it comes to determining the level of difficulties.Article Citation - WoS: 1Citation - Scopus: 1Modeling of Kappa Factor Using Multivariate Adaptive Regression Splines: Application To the Western Türkiye Ground Motion Dataset(Springer, 2024) Kurtulmus, Tevfik Ozgur; Yerlikaya-Ozkurt, Fatma; Askan, AysegulThe recent seismic activity on Turkiye's west coast, especially in the Aegean Sea region, shows that this region requires further attention. The region has significant seismic hazards because of its location in an active tectonic regime of North-South extension with multiple basin structures on soft soil deposits. Recently, despite being 70 km from the earthquake source, the Samos event (with a moment magnitude of 7.0 on October 30, 2020) caused significant localized damage and collapse in the Izmir city center due to a combination of basin effects and structural susceptibility. Despite this activity, research on site characterization and site response modeling, such as local velocity models and kappa estimates, remains sparse in this region. Kappa values display regional characteristics, necessitating the use of local kappa estimations from previous earthquake data in region-specific applications. Kappa estimates are multivariate and incorporate several characteristics such as magnitude and distance. In this study, we assess and predict the trend in mean kappa values using three-component strong-ground motion data from accelerometer sites with known VS30 values throughout western Turkiye. Multiple linear regression (MLR) and multivariate adaptive regression splines (MARS) were used to build the prediction models. The effects of epicentral distance Repi, magnitude Mw, and site class (VS30) were investigated, and the contributions of each parameter were examined using a large dataset containing recent seismic activity. The models were evaluated using well-known statistical accuracy criteria for kappa assessment. In all performance measures, the MARS model outperforms the MLR model across the selected sites.Article Citation - WoS: 6Citation - Scopus: 7A Simplified Method Based on Rssi Fingerprinting for Iot Device Localization in Smart Cities(Ieee-inst Electrical Electronics Engineers inc, 2024) Dogan, Deren; Dalveren, Yaser; Kara, Ali; Derawi, MohammadThe Internet of Things (IoT) has significantly improved location-based services in smart cities, such as automated public transportation and traffic management. Estimating the location of connected devices is a critical problem. Low Power Wide Area Network (LPWAN) technologies are used for localization due to their low power consumption and long communication range. Recent advances in Machine Learning have made Received Signal Strength Indicator (RSSI) fingerprinting with LPWAN technologies effective. However, this requires a connection between devices and gateways or base stations, which can increase network deployment, maintenance, and installation costs. This study proposes a cost-effective RSSI fingerprinting solution using IQRF technology for IoT device localization. The region of interest is divided into grids to provide training locations, and measurements are conducted to create a training dataset containing RSSI fingerprints. Pattern matching is performed to localize the device by comparing the fingerprint of the end device with the fingerprints in the created database. To evaluate the efficiency of the proposed solution, measurements were conducted in a short-range local area ( $80\times 30$ m) at 868 MHz. In the measurements, four IQRF nodes were utilized to receive the RSSIs from a transmitting IQRF node. The performances of well-known ML classifiers on the created dataset are then comparatively assessed in terms of test accuracy, prediction speed, and training time. According to the results, the Bagged Trees classifier demonstrated the highest accuracy with 96.87%. However, with an accuracy of 95.69%, the Weighted k-NN could also be a reasonable option for real-world implementations due to its faster prediction speed (37615 obs/s) and lower training time (28.1 s). To the best of the authors' knowledge, this is the first attempt to explore the feasibility of the IQRF networks to develop a RSSI fingerprinting-based IoT device localization in the literature. The promising results suggest that the proposed method could be used as a low-cost alternative for IoT device localization in short-range location-based smart city applications.Article Citation - WoS: 10Citation - Scopus: 16Student Engagement Research Trends of Past 10 Years: a Machine Learning-Based Analysis of 42,000 Research Articles(Springer, 2023) Gurcan, Fatih; Erdogdu, Fatih; Cagiltay, Nergiz Ercil; Cagiltay, KursatStudent engagement is critical for both academic achievement and learner satisfaction because it promotes successful learning outcomes. Despite its importance in various learning environments, research into the trends and themes of student engagement is scarce. In this regard, topic modeling, a machine learning technique, allows for the analysis of large amounts of content in any field. Thus, topic modeling provides a systematic methodology for identifying research themes, trends, and application areas in a comprehensive framework. In the literature, there is a lack of topic modeling-based studies that analyze the holistic landscape of student engagement research. Such research is important for identifying wide-ranging topics and trends in the field and guiding researchers and educators. Therefore, this study aimed to analyze student engagement research using a topic modeling approach and to reveal research interests and trends with their temporal development, thereby addressing a lack of research in this area. To this end, this study analyzed 42,517 peer-reviewed journal articles published from 2010 to 2019 using machine learning techniques. According to our findings, two new dimensions, "Community Engagement" and "School Engagement", were identified in addition to the existing ones. It is also envisaged that the next period of research and applications in student engagement will focus on the motivation-oriented tools and methods, dimensions of student engagement, such as social and behavioral engagement, and specific learning contexts such as English as a Foreign Language "EFL" and Science, Technology, Engineering and Math "STEM".Conference Object Citation - Scopus: 2Reinforcement Learning for Intrusion Detection(Springer Science and Business Media Deutschland GmbH, 2023) Saad,A.M.S.E.; Yildiz,B.Network-based technologies such as cloud computing, web services, and Internet of Things systems are becoming widely used due to their flexibility and preeminence. On the other hand, the exponential proliferation of network-based technologies exacerbated network security concerns. Intrusion takes an important share in the security concerns surrounding network-based technologies. Developing a robust intrusion detection system is crucial to solving the intrusion problem and ensuring the secure delivery of network-based technologies and services. In this paper, we propose a novel approach using deep reinforcement learning to detect intrusions to make network applications more secure, reliable, and efficient. As for the reinforcement learning approach, Deep Q-learning is used alongside a custom-built Gym environment that mimics network attacks and guides the learning process. The NSL-KDD dataset is used to create the reinforcement learning environment to train and evaluate the proposed model. The experimental results show that our proposed reinforcement learning approach outperforms other related solutions in the literature, achieving an accuracy that exceeds 93%. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.Master Thesis Makine Öğrenmesinde, Farklı Veri Temizleme Tekniklerlerinin Sonuç Ölçevleri Üzerindeki Etkisinin İncelenmesi(2022) Abbas, Israa Mustafa; Toker, SacipE-ticaret platformları ve çevrimiçi uygulamalar nedeniyle verilerin katlanarak büyümesi, veri analizi ve işlenmesi için büyük bir zorluk yarattı. Artık internetteki e-ticaret sitelerinin müşterilerinin satın aldıkları ürünler hakkında yorum yazmalarını sağlamak sık kullanılan bir uygulamadır. Bu incelemeler, bu ürünler hakkında değerli bilgi kaynakları sağlar. Bir ürün incelemesi, tüm çevrimiçi ürün şirketlerinde kullanılan duygusal analiz için önemli bir veri kaynağı içerir. Bu büyük miktarda veri etkisi büyük bir zorluk yaratır. Ancak, bu veri kümelerinin farklı veri sorunları vardır. Çoğu durumda, genellikle veriler yayınlanmadan önce çeşitli veri madenciliği teknikleri kullanılır. Mekansal olarak, görünmeyen verileri tahmin etmek için geçmiş ve etiketlenmiş veriler üzerinde eğitilen denetimli makine öğrenimi modellerinde, modelin daha önce öğrenmediği veriler. Bu tezde ayrıca makine öğrenmesinde deney çalışması tasarımına odaklandık. [1]. Bir sebep-sonuç ilişkisi bulmak için düzenli olarak Ronald Fisher'ın teorilerini [2] uygularız. Bu deneysel çalışma tasarımını uygulamak için, doğal dil işleme (NLP) yaklaşımı olan duygusal analiz ile denetimli makine öğrenmesi sınıflandırma algoritmalarını seçtik. Kuruluşların bir ürün veya hizmet hakkındaki görüşleri tanımlaması ve kategorilere ayırması için ortak bir yol. Duyguları ve öznel bilgileri elde etmek için metin madenciliği yapmak için veri madenciliği, makine öğrenimi ve yapay zeka kullanmayı içerir [3]. Bu çalışma, beş deney grubunun (yinelenen veri, noktalama işaretleri, durdurma sözcükleri, limmatezr, TF-IDF transform) etkisini analiz etmek ve bunları bir kontrol grubuyla karşılaştırmak (veri temizleme işlemi yapılmamış) için Multinominal Naïve Bays, Random Forest ve Lojistik Regresyon ile kurulmuştur. Uygulamalı. Deney grubunun üç modelin verimliliğine ve sınıflandırma oranına etkisini belirlemek ve ilginç gözlemleri açıklamak. Simülasyonlar, yirmi dört farklı kategoriden Amazon Product Review veri kümesinden rastgele seçilen 353 proje üzerinde çalıştırıldı. Böylece, veri seti Amazon.com'dan McAuley ve Leskovec [4][5] tarafından toplanmıştır. Metrik veri seti toplandıktan sonra analiz için SPSS yazılımı kullanılmıştır. Bu araştırma sorusunu ve kullanılan ölçeğin tanımlayıcı istatistiklerini incelemek için tekrarlı ölçüm ANOVA yapılmıştır. Analizin sonucu, veri temizlemenin makine öğrenimi modellerinin performansı üzerinde farklı bir etkisinin olduğunu göstermektedir. Aynı durumlarda rasgele ormanda olumlu, çok taraflı naif koylarda ve lojistik regresyonda olumsuz etkilenir. Diğer durumlarda, hiçbir etkisi olmadı. Genel olarak, deneysel sonuçlar Random Forest sınıflandırıcısının, Multinominal Naive Bayes sınıflandırıcısına ve Logistic Regression sınıflandırıcısına göre veri temizlemeye daha duyarlı olduğunu ve iki algoritmanın da temiz olmayan veri setinde yüksek bir sınıflandırma puanı elde ettiğini göstermiştir. Ayrıca, deney sonuçları, veri sorunları davranışının makine öğrenimi modelinde farklılık gösterdiğini gösterdi. Tüm makine öğrenimi algoritmalarında veri kalitesi sorunlarını alakasız veriler olarak kabul edemeyiz.Master Thesis Kolektif Derin Öğrenme ve Transfer Öğrenme Yoluyla Mahsul ve Meyve Sınıflandırması(2023) Daşkın, Zeynep Dilan; Khan, Muhammad UmerSon yıllarda yapılan teknolojik gelişmeler, tarım sektörünü hızlı bir şekilde yeniden şekillendirerek, daha önce kullanılmakta olan geleneksel metotlarda devrim yaratıyor ve insanlık için daha sürdürülebilir ve üretken bir geleceğin yolunu açıyor. Tarım sektörü, makine öğrenmesi, sensor ve mekanizmaları kullanarak otomatikleşirken, aynı zamanda verimlilik artımı, kaynak yönetimi ve mahsul sağlığı açısından da köklü bir değişim yaşıyor. Bu tez çalışmasında, son teknoloji makine öğrenimi tabanlı görüntü işleme teknikleri ve algoritmaları kapsamlı bir şekilde araştırılmış ve analiz edilmiştir. Amaç, çeşitli mahsulleri, meyveleri ve sebzeleri doğru bir şekilde tespit eden, tanımlayan ve sınıflandıran sağlam metodolojiler geliştirmektir. Nihai hedef, gelişen tarımsal otomasyon endüstrisine önemli ölçüde katkıda bulunmak, süreçleri kolaylaştırmak ve tarım sektöründe verimliliği arttırmaktır. Gerçek yaşam koşullarına en yakın sonuçları elde etmek için, yazarların kendi oluşturduğu mahsul veri kümeleri bu araştırma boyunca önerilen algoritmalara entegre edilmiş ve kullanılmıştır. Bu çalışma sırasında kullanılan mevcut Evrişimsel Sinir Ağları algoritmaları AlexNet, GoogleNet, InceptionV3, SqueezeNet, DenseNet ve VGG-16'dır. Bu çalışmada, doğruluk, kayıp, F1-skoru, tahmin, kesinlik ve duyarlılık olmak üzere genel değerlendirme metriklerinin performansını yükseltmek için çeşitli gelişmiş algoritmalar araştırılmış ve incelenmiştir. Özellikle, tarımsal otomasyon sisteminin etkinliğini ve güvenilirliğini arttırmayı amaçlayan Kolektif Öğrenme ve Öğrenme Aktarımı adlı iki metot tanıtılmış ve kapsamlı bir şekilde analiz edilmiştir. Çalışmanın sonuçları, önerilen algoritmaların tarım endüstrisinde son derece etkili olduğunu ve istenen sonuçları benzersiz bir doğruluk ve hassasiyetle sunma yeteneklerini kanıtlamaktadır. Bu sonuçlar, bu algoritmaların operasyonel verimliliği önemli ölçüde arttırma, kaynak tahsisini optimize etme ve gelecek için tarımsal otomasyonda sürdürülebilir uygulamaları teşvik etme gibi potansiyellerini de göstermekte ve teyit etmektedir.Article Citation - WoS: 14Citation - Scopus: 17Enhancement of Quality and Quantity of Woody Biomass Produced in Forests Using Machine Learning Algorithms(Pergamon-elsevier Science Ltd, 2023) Peng, Wei; Sadaghiani, Omid KarimiForest is considered a significant source of woody biomass production. Sustainable production of wood, lower emittance of CO2 from burning, and lower amount of sulfur and heavy metals are the advantages of wood rather than fossil fuels. The quality and quantity of woody biomass production are a function of some operations including genetic modifications, high-quality forestry, evaluation, monitoring, storage, and transportation. Due to surveying numerous related works, it was found that there is a considerable reviewing gap in analyzing and collecting the applications of Machine Learning in the quality and quantity of woody biomass. To fill this gap in the current work, the above-mentioned operations are explained followed by the applications of Machine Learning algorithms. Conclusively, Machine Learning and Deep Learning can be employed in estimating main effective factors on trees growth, classification of seeds, trees, and regions, as well as providing decision-making tools for farmers or governors, evaluation of biomass, understanding the relation between the woody bimass internal structure and bio-fuel production, the ultimate and proximate analyses, prediction of wood contents and dimensions, determination of the proportion of mixed woody materials, monitoring for early disease identifi-cation and classification, classifying trees diseases, estimating evapotranspiration, collecting information about forest regions and its quality, nitrogen concentration in trees, choosing viable storage sites for storage depots and improving the solution, classifying different filling levels in silage, estimating acetic acid synthesis and aerobic reactions in silage, determining crop quantity in silo, estimating the methane production, and monitoring and predicting water content, quality and quantity of stored biomass, forecasting the demand, path way and on-time performance predicting, truck traffic predicting, and behavioral analysis and facility planning.Article Citation - WoS: 6Citation - Scopus: 10Beyond Rouge: a Comprehensive Evaluation Metric for Abstractive Summarization Leveraging Similarity, Entailment, and Acceptability(World Scientific Publ Co Pte Ltd, 2024) Briman, Mohammed Khalid Hilmi; Yıldız, Beytullah; Yildiz, Beytullah; Yıldız, BeytullahA vast amount of textual information on the internet has amplified the importance of text summarization models. Abstractive summarization generates original words and sentences that may not exist in the source document to be summarized. Such abstractive models may suffer from shortcomings such as linguistic acceptability and hallucinations. Recall-Oriented Understudy for Gisting Evaluation (ROUGE) is a metric commonly used to evaluate abstractive summarization models. However, due to its n-gram-based approach, it ignores several critical linguistic aspects. In this work, we propose Similarity, Entailment, and Acceptability Score (SEAScore), an automatic evaluation metric for evaluating abstractive text summarization models using the power of state-of-the-art pre-trained language models. SEAScore comprises three language models (LMs) that extract meaningful linguistic features from candidate and reference summaries and a weighted sum aggregator that computes an evaluation score. Experimental results show that our LM-based SEAScore metric correlates better with human judgment than standard evaluation metrics such as ROUGE-N and BERTScore.
