Scopus İndeksli Yayınlar Koleksiyonu
Permanent URI for this collectionhttps://hdl.handle.net/20.500.12416/8651
Browse
24 results
Search Results
Article Citation - WoS: 2Citation - Scopus: 2Comparative Evaluation of YOLO and Gemini AI Models for Road Damage Detection and Mapping(MDPI, 2025) Demirel, Zeynep; Nasraldeen, Shvan Tahir; Pehlivan, Oyku; Shoman, Sarmad; Albdairi, Mustafa; Almusawi, AliEfficient detection of road surface defects is vital for timely maintenance and traffic safety. This study introduces a novel AI-powered web framework, TriRoad AI, that integrates multiple versions of the You Only Look Once (YOLO) object detection algorithms-specifically YOLOv8 and YOLOv11-for automated detection of potholes and cracks. A user-friendly browser interface was developed to enable real-time image analysis, confidence-based prediction filtering, and severity-based geolocation mapping using OpenStreetMap. Experimental evaluation was conducted using two datasets: one from online sources and another from field-collected images in Ankara, Turkey. YOLOv8 achieved a mean accuracy of 88.43% on internet-sourced images, while YOLOv11-B demonstrated higher robustness in challenging field environments with a detection accuracy of 46.15%, and YOLOv8 followed closely with 44.92% on mixed field images. The Gemini AI model, although highly effective in controlled environments (97.64% detection accuracy), exhibited a significant performance drop of up to 80% in complex field scenarios, with its accuracy falling to 18.50%. The proposed platform's uniqueness lies in its fully integrated, browser-based design, requiring no device-specific installation, and its incorporation of severity classification with interactive geospatial visualization. These contributions address current gaps in generalization, accessibility, and practical deployment, offering a scalable solution for smart infrastructure monitoring and preventive maintenance planning in urban environments.Article Detection and Classification of Femoral Neck Fractures From Plain Pelvic X-Rays Using Deep Learning and Machine Learning Methods(Turkish Assoc Trauma Emergency Surgery, 2025) Sevinc, Huseyin Fatih; Ureten, Kemal; Karadeniz, Talha; Gultekin, Gokhan KorayBackground: Femoral neck fractures are a serious health concern, particularly among the elderly. The aim of this study is to diagnose and classify femoral neck fractures from plain pelvic X-rays using deep learning and machine learning algorithms, and to compare the performance of these methods. Methods: The study was conducted on a total of 598 plain pelvic X-ray images, including 296 patients with femoral neck fractures and 302 individuals without femoral neck fractures. Initially, transfer learning was applied using pre-trained deep learning models: VGG-16, ResNet-50, and MobileNetv2. Results: The pre-trained VGG-16 network demonstrated slightly better performance than ResNet-50 and MobileNetV2 for detecting and classifying femoral neck fractures. Using the VGG-16 model, the following results were obtained: 95.6% accuracy, 95.5% sensitivity, 93.3% specificity, 95.7% precision, 95.5% F1 Score, a Cohen's kappa of 0.91, and the Receiver Operating Characteristic (ROC) curve of 0.99. Subsequently, features extracted from the convolution layers of VGG-16 were classified using common machine learning algorithms. Among these, the k-nearest neighbor (k-NN) algorithm outperformed the others and exceeded the accuracy of the VGG-16 model by 1%. Conclusion: Successful results were obtained using deep learning and machine learning methods for the detection and classification of femoral neck fractures. The model can be further improved through multi-center studies. The proposed model may be especially useful for physicians working in emergency departments and for those not having sufficient experience in evaluating plain pelvic radiographs.Article Citation - WoS: 1Citation - Scopus: 1Ordered Clustering-Based Semantic Music Recommender System Using Deep Learning Selection(Tech Science Press, 2025) Ha, Weitao; Gang, Sheng; Navaei, Yahya D.; Gezawa, Abubakar S.; Nanehkaran, Yaser A.Music recommendation systems are essential due to the vast amount of music available on streaming platforms, which can overwhelm users trying to find new tracks that match their preferences. These systems analyze users' emotional responses, listening habits, and personal preferences to provide personalized suggestions. A significant challenge they face is the "cold start" problem, where new users have no past interactions to guide recommendations. To improve user experience, these systems aim to effectively recommend music even to such users by considering their listening behavior and music popularity. This paper introduces a novel music recommendation system that combines order clustering and a convolutional neural network, utilizing user comments and rankings as input. Initially, the system organizes users into clusters based on semantic similarity, followed by the utilization of their rating similarities as input for the convolutional neural network. This network then predicts ratings for unreviewed music by users. Additionally, the system analyses user music listening behaviour and music popularity. Music popularity can help to address cold start users as well. Finally, the proposed method recommends unreviewed music based on predicted high rankings and popularity, taking into account each user's music listening habits. The proposed method combines predicted high rankings and popularity by first selecting popular unreviewed music that the model predicts to have the highest ratings for each user. Among these, the most popular tracks are prioritized, defined by metrics such as frequency of listening across users. The number of recommended tracks is aligned with each user's typical listening rate. The experimental findings demonstrate that the new method outperformed other classification techniques and prior recommendation systems, yielding a mean absolute error (MAE) rate and root mean square error (RMSE) rate of approximately 0.0017, a hit rate of 82.45%, an average normalized discounted cumulative gain (nDCG) of 82.3%, and a prediction accuracy of new ratings at 99.388%.Article Citation - WoS: 11Citation - Scopus: 17Convolutional Neural Network-Based Deep Learning for Landslide Susceptibility Mapping in the Bakhtegan Watershed(Nature Portfolio, 2025) Feng, Li; Zhang, Maosheng; Mao, Yimin; Liu, Hao; Yang, Chuanbo; Dong, Ying; Nanehkaran, Yaser A.Landslides pose a significant threat to infrastructure, ecosystems, and human safety, necessitating accurate and efficient susceptibility assessment methods. Traditional models often struggle to capture the complex spatial dependencies and interactions between geological and environmental factors. To address this gap, this study employs a deep learning approach, utilizing a convolutional neural network (CNN) for high-precision landslide susceptibility mapping in the Bakhtegan watershed, southwestern Iran. A comprehensive landslide inventory was compiled using 235 documented landslide locations, validated through remote sensing and field surveys. An equal number of non-landslide locations were systematically selected to ensure balanced model training. Fifteen key conditioning factors-including topographical, geological, hydrological, and climatological variables-were incorporated into the model. While traditional statistical methods often fail to extract spatial hierarchies, the CNN model effectively processes multi-dimensional geospatial data, learning intricate patterns influencing slope instability. The CNN model outperformed other classification approaches, achieving an accuracy of 95.76% and a precision of 95.11%. Additionally, error metrics confirmed its reliability, with a mean absolute error (MAE) of 0.11864, mean squared error (MSE) of 0.18796, and root mean squared error (RMSE) of 0.18632. The results indicate that the northern and northeastern regions of the Bakhtegan watershed are highly susceptible to landslides, highlighting areas where proactive mitigation strategies are crucial. This study demonstrates that deep learning, particularly CNNs, offers a powerful and scalable solution for landslide susceptibility assessment. The findings provide valuable insights for urban planners, engineers, and policymakers to implement effective risk reduction strategies and enhance resilience in landslide-prone regions.Conference Object Citation - WoS: 10Citation - Scopus: 21Multi-Label Classification of Text Documents Using Deep Learning(Ieee, 2020) Mohammed, Hamza Haruna; Dogdu, Erdogan; Gorur, Abdul Kadir; Choupani, RoyaRecently, studies in the field of Natural Language Processing and its related applications continue to mount up. Machine learning is proven to be predominantly data-driven in the sense that generic model building methods are used and then tailored to specific application domains. Needless to say, this has proven to be a very effective approach in modeling the complicated data dependencies we frequently experience in practice, making very few assumptions, and allowing the information to talk for themselves. Examples of these applications can be found in chemical process engineering, climate science, healthcare, and linguistic processing systems for natural languages, to name a few. Text classification is one of the important machine learning tasks that is used in many digital applications today; such as in document filtering, search engines, document management systems, and many more. Text classification is the process of categorizing of text documents into a given set of labels. Furthermore, multi-label text classification is the task of categorization of text documents into one or more labels simultaneously. Over the years, many methods for classifying text documents have been proposed, including the popularly known bag of words (BoW) method, support vector machine (SVM), tree induction, and label-vector embedding, to mention a few. These kinds of tools can be used in many digital applications, such as document filtering, search engines, document management systems, etc. Lately, deep learning-based approaches are getting more attention, especially in extreme multi-label text classification case. Deep learning has proven to be one of the major solutions to many machine learning applications, especially those involving high-dimensional and unstructured data. However, it is of paramount importance in many applications to be able to reason accurately about the uncertainties associated with the predictions of the models. In this paper, we explore and compare the recent deep learning-based methods for multi-label text classification. We investigate two scenarios. First, multi-label classification model with ordinary embedding layer, and second with Glove, word2vec, and FastText as pre-trained embedding corpus for the given models. We evaluated these different neural network model performances in terms of multi-label evaluation metrics for the two approaches, and compare the results with the previous studies.Article Enhancing Session-Based Trip Recommendations Using Matrix Factorization: a Study on Algorithm Efficiency and Resource Utilization(Springer, 2025) Mat, Abdullah Ugur; Saran, Ayse NurdanAs the impact and usefulness of recommendation systems continue to grow, their importance becomes more and more pronounced. Therefore, it is crucial to design and implement recommendation systems that are both efficient and highly accurate to meet the increasing demands and expectations. This study focuses on a model awarded first place in a travel forecasting recommendation system competition. This study aims to enhance matrix factorization-based recommender systems by conducting a comprehensive analysis of various factors. This includes examining the effects of resource utilization and recurrent neural network (RNN) algorithms on session-based factorization, as well as evaluating the influence of embeddings and optimization techniques concerning their efficiency and accuracy. The gated recurrent unit (GRU) algorithm has produced more accurate results for reduced datasets than long short-term memory (LSTM). Some modifications have been made on the embedding layers, and the results have been observed. In addition, the model's optimizer is changed, and the performance of different optimizers is evaluated. While random reduction of the dataset has led to a decrease in the success rate, methodical reduction has significantly increased the success rate. The highest and most reliable success rate (0.6654) was achieved by applying the selection method, which reduced the dataset to 1 M records from 1.5 M records. Optimizers have shown a wide range of effects on hardware.Article Citation - WoS: 15Citation - Scopus: 15Deep Learning Method for Compressive Strength Prediction for Lightweight Concrete(Techno-press, 2023) Nanehkaran, Yaser A.; Azarafza, Mohammad; Pusatli, Tolga; Bonab, Masoud Hajialilue; Irani, Arash Esmatkhah; Kouhdarag, Mehdi; Derakhshani, RezaConcrete is the most widely used building material, with various types including high-and ultra-high-strength, reinforced, normal, and lightweight concretes. However, accurately predicting concrete properties is challenging due to the geotechnical design code's requirement for specific characteristics. To overcome this issue, researchers have turned to new technologies like machine learning to develop proper methodologies for concrete specification. In this study, we propose a highly accurate deep learning-based predictive model to investigate the compressive strength (UCS) of lightweight concrete with natural aggregates (pumice). Our model was implemented on a database containing 249 experimental records and revealed that water, cement, water-cement ratio, fine-coarse aggregate, aggregate substitution rate, fine aggregate replacement, and superplasticizer are the most influential covariates on UCS. To validate our model, we trained and tested it on random subsets of the database, and its performance was evaluated using a confusion matrix and receiver operating characteristic (ROC) overall accuracy. The proposed model was compared with widely known machine learning methods such as MLP, SVM, and DT classifiers to assess its capability. In addition, the model was tested on 25 laboratory UCS tests to evaluate its predictability. Our findings showed that the proposed model achieved the highest accuracy (accuracy=0.97, precision=0.97) and the lowest error rate with a high learning rate (R2=0.914), as confirmed by ROC (AUC=0.971), which is higher than other classifiers. Therefore, the proposed method demonstrates a high level of performance and capability for UCS predictions.Article Citation - WoS: 18Citation - Scopus: 27Application of Bilstm-Crf Model With Different Embeddings for Product Name Extraction in Unstructured Turkish Text(Springer London Ltd, 2024) Arslan, SerdarNamed entity recognition (NER) plays a pivotal role in Natural Language Processing by identifying and classifying entities within textual data. While NER methodologies have seen significant advancements, driven by pretrained word embeddings and deep neural networks, the majority of these studies have focused on text with well-defined grammar and structure. A significant research gap exists concerning NER in informal or unstructured text, where traditional grammar rules and sentence structure are absent. This research addresses this crucial gap by focusing on the detection of product names within unstructured Turkish text. To accomplish this, we propose a deep learning-based NER model which combines a Bidirectional Long Short-Term Memory (BiLSTM) architecture with a Conditional Random Field (CRF) layer, further enhanced by FastText embeddings. To comprehensively evaluate and compare our model's performance, we explore different embedding approaches, including Word2Vec and Glove, in conjunction with the Bidirectional Long Short-Term Memory and Conditional Random Field (BiLSTM-CRF) model. Furthermore, we conduct comparisons against BERT to assess the efficacy of our approach. Our experimentation utilizes a Turkish e-commerce dataset gathered from the internet, where traditional grammatical and structural rules may not apply. The BiLSTM-CRF model with FastText embeddings achieved an F1 score value of 57.40%, a precision value of 55.78%, and a recall value of 59.12%. These results indicate promising performance in outperforming other baseline techniques. This research contributes to the field of NER by addressing the unique challenges posed by unstructured Turkish text and opens avenues for improved entity recognition in informal language settings, with potential applications across various domains.Article Citation - WoS: 12Citation - Scopus: 16A Metaheuristic-Guided Machine Learning Approach for Concrete Strength Prediction With High Mix Design Variability Using Ultrasonic Pulse Velocity Data(Elsevier, 2023) Selcuk, S.; Tang, P.Assessment of concrete strength in existing structures is a common engineering problem. Several attempts in the literature showed the potential of ML methods for predicting concrete strength using concrete properties and NDT values as inputs. However, almost all such ML efforts based on NDT data trained models to predict concrete strength for a specific concrete mix design. We trained a global ML-based model that can predict concrete strength for a wide range of concrete types. This study uses data with high variability for training a metaheuristic-guided ANN model that can cover most concrete mixes used in practice. We put together a dataset that has large variations of mix design components. Training an ANN model using this dataset introduced significant test errors as expected. We optimized hyperparameters, architecture of the ANN model and performed feature selection using genetic algorithm. The proposed model reduces test errors from 9.3 MPa to 4.8 MPa.Article Citation - WoS: 11Citation - Scopus: 12Diagnosis of Osteoarthritic Changes, Loss of Cervical Lordosis, and Disc Space Narrowing on Cervical Radiographs With Deep Learning Methods(Turkish Joint Diseases Foundation, 2022) Tokdemir, Gul; Ureten, Kemal; Atalar, Ebru; Duran, Semra; Maras, Hakan; Maras, YukselObjectives: In this study, we aimed to differentiate normal cervical graphs and graphs of diseases that cause mechanical neck pain by using deep convolutional neural networks (DCNN) technology. Materials and methods: In this retrospective study, the convolutional neural networks were used and transfer learning method was applied with the pre-trained VGG-16, VGG-19, Resnet-101, and DenseNet-201 networks. Our data set consisted of 161 normal lateral cervical radiographs and 170 lateral cervical radiographs with osteoarthritis and cervical degenerative disc disease. Results: We compared the performances of the classification models in terms of performance metrics such as accuracy,
- «
- 1 (current)
- 2
- 3
- »