Browsing by Author "Kurt, Zühal"

Now showing 1 - 12 of 12

Citation Count: 1
Deep Learning-Based COVID-19 Detection Using Lung Parenchyma CT Scans
(Springer international Publishing Ag, 2022) Kaya, Zeynep; Kurt, Zuhal; Koca, Nizameddin; Cicek, Sumeyye; Isik, Sahin; Computer Engineering
During the outbreak of the COVID-19 pandemic, it is important to improve early diagnosis using effective ways in order to lower the risks and further spread of the viruses as early as possible. This is also important when it comes to appropriate treatments and the reduction of mortality rates. In this respect, computer tomography (CT) scanning is a useful technique in detecting COVID-19. The present paper, as such, is an attempt to contribute to this process by generating an open-source, CT-based image dataset. This dataset contains the CT scans of lung parenchyma regions of 180 COVID-19 positives and 86 COVID-19 negative patients, all from Bursa Yuksek Ihtisas Training and Research Hospital. The experimental studies demonstrate that this dataset is effectively utilized deep learning-based models for diagnostic purposes. Firstly, a smart segmentation mechanism based on the k-means algorithm is applied to this dataset as a pre-processing stage. Then, the performance of the proposed method is evaluated using InceptionV3 and Xception convolutional neural networks, yielding a 96.20% and 96.55% accuracy rate and 95.00% and 95.50% F1-score, respectively. These state-of-the-art models are observed to detect COVID-19 cases faster and more accurately. In addition, the fine-tuning stage of the convolutional neural network (CNN) features sufficiently improves this accuracy rate. For these features, the support vector machine (SVM) classifier is used, resulting in remarkable 96.76% accuracy rate and 95.81% F1-score. The implications of the proposed method are immense both for present-day applications as well as future developments.
Citation Count: 0
Development of 'SeniorHouse' Application for Online Education
(Ieee, 2021) Kurt, Zuhal; Keskin, Ahmet; Kocaova, Gokhan; Boynukisa, Levent; Ulgen, Okan Can; Computer Engineering
With the diffusion of technology worldwide, almost every individual is starting to have at least one mobile device, making mobile applications very popular and an important sector. Given the direction in which many sectors, including education, are moving, there is a clear and evident need for communication on-line rather than in-person. Concerning universities, the communication gap between teachers and students was particularly compromised in light of the Covid-19 pandemic. This study is aimed to fill in this gap by developing a social platform where students and teachers can share their posts and create their personal profiles. The Firebase database and Android Studio are the chosen technologies for this paper due to their compatibility as well as popularity as a chat application, along with Java as the programming language. The developed platform, named "SeniorHouse", is focused on educational content to provide both faculty and students with a two-way, real-time communication to share updates, posts, documents, project ideas, and personal profiles, to name a few. The SeniorHouse application has significant potentials to be expanded into a major hub for future on-line education.
Citation Count: 5
Evaluation of EfficientNet models for COVID-19 detection using lung parenchyma
(Springer London Ltd, 2023) Kurt, Zuhal; Isik, Sahin; Kaya, Zeynep; Anagun, Yildiray; Koca, Nizameddin; Cicek, Suemeyye; Computer Engineering
When the COVID-19 pandemic broke out in the beginning of 2020, it became crucial to enhance early diagnosis with efficient means to reduce dangers and future spread of the viruses as soon as possible. Finding effective treatments and lowering mortality rates is now more important than ever. Scanning with a computer tomography (CT) scanner is a helpful method for detecting COVID-19 in this regard. The present paper, as such, is an attempt to contribute to this process by generating an open-source, CT-based image dataset. This dataset contains the CT scans of lung parenchyma regions of 180 COVID-19-positive and 86 COVID-19-negative patients taken at the Bursa Yuksek Ihtisas Training and Research Hospital. The experimental studies show that the modified EfficientNet-ap-nish method uses this dataset effectively for diagnostic purposes. Firstly, a smart segmentation mechanism based on the k-means algorithm is applied to this dataset as a preprocessing stage. Then, performance pretrained models are analyzed using different CNN architectures and with our Nish activation function. The statistical rates are obtained by the various EfficientNet models and the highest detection score is obtained with the EfficientNet-B4-ap-nish version, which provides a 97.93% accuracy rate and a 97.33% F1-score. The implications of the proposed method are immense both for present-day applications and future developments.
Citation Count: 1
A Graph-Based Recommendation Algorithm on Quaternion Algebra
(Springer, 2022) Kurt,Z.; Gerek,Ö.N.; Bilge,A.; Özkan,K.; Computer Engineering
This study presents a novel Quaternion-based link prediction method to be used in different recommendation systems. The method performs Quaternion algebra-based computations while making use of expressive and wide-ranged learning properties of the Hamilton products. The proposed key capabilities rely on link prediction to boost performance in top-N recommendation tasks. According to the achieved experimental results, the proposed method allows for highly improved performance according to three quality measurements: (i) hits rate, (ii) coverage, and (iii) novelty; when applied to two datasets, namely the Movielens and Hetrec datasets. To assess the flexibility level of the proposed algorithm in terms of incorporating alternative sources of information, further wide-scale tests are carried out on three subsets of the Amazon dataset. Hence, the effectiveness of Quaternion algebra in graph-based recommendation algorithms is verified. The algorithms suggested here are further enhanced using similarity and dissimilarity factors between users and items, as well as ‘like’ and ‘dislike’ relationships between users and items. It is observed that this approach is adaptable by incorporating different information sources and can successfully overcome the drawbacks of conventional graph-based recommender systems. It is argued that the proposed novel idea of Quaternion-based link prediction method stands as a superior alternative to existing methods. © 2022, The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd.
Citation Count: 0
Güvenlik Sistemleri için Silah ve Bıçak Tanıma
(2021) Işık, Şahin; Özkan, Şerif Ercan; Kurt, Zuhal; Computer Engineering
Bu çalışma, halka açık yerlerde güvenlik sorunlarının üstesinden gelmek için etkili ve yenilikçi bir çözüm sunmaktadır. Alternatif bir video gözetim sistemi olarak, önerilen yöntem videolardan silah ve bıçak nesnelerini gerçek zamanlı olarak algılar ve yerelleştirir. Evrişimsel Sinir Ağı tabanlı nesne algılama ile bağlantılı olarak, en yüksek performansa sahip silah ve bıçak nesnelerini tespit etmek için Hızlı-Bölgesel Tabanlı Evrişimsel Sinir Ağı yapısı uygulanmıştır. Test görüntüleri üzerinde simülasyon gerçekleştirdikten sonra, geliştirilen sistemin F1-skor performansı yaklaşık %70 tanıma oranı olarak elde edilmiştir. Eğitilen Faster R-CNN modeli, uçak, otobüs durağı, stadyum ve güvenliğin önemli bir faktör olduğu kamu taşıtları da dâhil olmak üzere farklı halka açık yerler için kullanılabilir. Ayrıca, geliştirilen yöntem, tehlikeli nesnelerin raporlanması ve bu tür nesnelerin neden olduğu risklerin en aza indirilmesi açısından yerel gözetim sistemine gömülebilir.
Citation Count: 0
Insurance Sales Forecast Using Machine Learning Algorithms
(Springer international Publishing Ag, 2022) Kurt, Zuhal; Varyok, Emrecan; Ayhan, Ege Baran; Bilgin, Mehmet Turhan; Duru, Duygu; Computer Engineering
Car accidents and the possible resulting loss of assets or life are issues for every car owner that must contend with some point in their driving life. Driving is an inherently dangerous act, even if it does not seem so at first, resulting in greater than 33,000 fatal vehi le crashes in USA in 2019 alone. However, the loss of life and possible damages can be reduced with the help of insurances. Insurance is an arrangement under which a person or agency receives financial security or reimbursement from an insurance provider in the form of a policy. Insurances help limit the losses of the customers when an undesirable event occurs, such as a car crash or a heart attack. Vehicle insurance provides customers monetary compensation after unfortunate accidents, provided they annually pay premium fees to the companies first. Our goal is to develop a machine learning algorithm that predicts customers who are interested in getting or renewing their vehicle insurance with the help of personal, vehicle, contact, and previous insurance data. The insurance sales forecast is helpful to companies, since they can then accordingly plan its communication strategy to reach out to those customers and optimize its business model and revenue, while also being beneficial to customers, who can go through the process and the aftermath of car accidents easier thanks to their monetary compensation. In this paper, the Health Insurance Cross-Sell Prediction dataset is used. The proposed model tries getting the value by training itself on a train and test dataset and will result in a categorical response feature based on the aforementioned data with the aid of well-known machine learning algorithms: k-nearest neighbors, random forest, support vector machines, Naive Bayes, and logistic regression.
Citation Count: 2
A Multi Source Graph-Based Hybrid Recommendation Algorithm
(Springer Science and Business Media Deutschland GmbH, 2021) Kurt,Z.; Gerek,Ö.N.; Bilge,A.; Özkan,K.; Computer Engineering
Images that widely exist on e-commerce sites, social networks, and many other applications are one of the most important information resources integrated into the recently deployed image-based recommender systems. In the latest studies, researchers have jointly considered ratings and images to generate recommendations, many of which are still restricted to limited information sources, sources namely, ratings with another input data, or which require the pre-existence of domain knowledge to generate recommendations. In this paper, a new graph-based hybrid framework is introduced to generate recommendations and overcome these challenges. Firstly, a simple overview of the framework is provided and, then, two different information sources (visual images and numerical ratings) are utilized to describe how the proposed framework can be developed in practice. Furthermore, the users’ visual preferences are determined based on which item they have already purchased. Then, each user is represented as a visual feature vector. Finally, the similarity factors between the users or items are evaluated from the user visual-feature or item visual-feature matrices, to be included the proposed algorithm for more efficiency. The proposed hybrid recommendation method depends on a link prediction approach and reveals the significant potential for performance improvement in top-N recommendation tasks. The experimental results demonstrate the superior performance of the proposed appraoch using three quality measurements - hit-ratio, recall, and precision - on the three subsets of the Amazon dataset, as well as its flexibility to incorporate different information sources. Finally, it is concluded that hybrid recommendation algorithms that use the integration of multiple types of input data perform better than previous recommendation algorithms that only utilize one type of input data. © 2021, The Author(s), under exclusive license to Springer Nature Switzerland AG.
Citation Count: 1
Recurrent Neural Networks for Spam E-mail Classification on an Agglutinative Language
(2020) Işık, Şahin; Kurt, Zuhal; Anagun, Yildiray; Özkan, Kemal; Computer Engineering
In this study, we have provided an alternative solution to spam and legitimate email classification problem. The different deep learning architectures are applied on two feature selection methods, including the Mutual Information (MI) and Weighted Mutual Information (WMI). Firstly, feature selection methods including WMI and MI are applied to reduce number of selected terms. Secondly, the feature vectors are constructed with concept of the bag-of-words (BoW) model. Finally, the performance of system is analyzed with using Artificial Neural Network (ANN), Long Short-Term Memory (LSTM) and Bidirectional Long Short-Term Memory (BILSTM) models. After experimental simulations, we have observed that there is a competition between detection results of using WMI and MI when commented with accuracy rates for the agglutinative language, namely Turkish. The experimental scores show that the LSTM and BILSTM give 100% accuracy scores when combined with MI or WMI, for spam and legitimate emails. However, for particular cross validation, the performance WMI is higher than MI features in terms e-mail grouping. It turns out that WMI and MI with deep learning architectures seem more robust to spam email detection when considering the high detection scores.
Citation Count: 7
Recurrent neural networks for spam E-mail classification on an agglutinative language
(Ismail Saritas, 2020) Işik,S.; Kurt,Z.; Anagun,Y.; Ozkan,K.; Computer Engineering
In this study, we have provided an alternative solution to spam and legitimate email classification problem. The different deep learning architectures are applied on two feature selection methods, including the Mutual Information (MI) and Weighted Mutual Information (WMI). Firstly, feature selection methods including WMI and MI are applied to reduce number of selected terms. Secondly, the feature vectors are constructed with concept of the bag-of-words (BoW) model. Finally, the performance of system is analyzed with using Artificial Neural Network (ANN), Long Short-Term Memory (LSTM) and Bidirectional Long Short-Term Memory (BILSTM) models. After experimental simulations, we have observed that there is a competition between detection results of using WMI and MI when commented with accuracy rates for the agglutinative language, namely Turkish. The experimental scores show that the LSTM and BILSTM give 100% accuracy scores when combined with MI or WMI, for spam and legitimate emails. However, for particular cross-validation, the performance WMI is higher than MI features in terms e-mail grouping. It turns out that WMI and MI with deep learning architectures seem more robust to spam email detection when considering the high detection scores. © 2020, Ismail Saritas. All rights reserved.
Citation Count: 1
Similarity-inclusive Link Prediction with Quaternions
(Science and Technology Publications, Lda, 2021) Kurt,Z.; Gerek,Ö.N.; Bilge,A.; Özkan,K.; Computer Engineering
This paper proposes a Quaternion-based link prediction method, a novel representation learning method for recommendation purposes. The proposed algorithm depends on and computation with Quaternion algebra, benefiting from the expressiveness and rich representation learning capability of the Hamilton products. The proposed method depends on a link prediction approach and reveals the significant potential for performance improvement in top-N recommendation tasks. The experimental results indicate the superior performance of the approach using two quality measurements – hits rate, and coverage - on the Movielens and Hetrec datasets. Additionally, extensive experiments are conducted on three subsets of the Amazon dataset to understand the flexibility of this algorithm to incorporate different information sources and demonstrate the effectiveness of Quaternion algebra in graph-based recommendation algorithms. The proposed algorithms obtain comparatively higher performance, they are improved with similarity factors. The results show that the proposed quaternion-based algorithm can effectively deal with the deficiencies in graph-based recommender system, making it a preferable alternative among the other available methods. Copyright © 2021 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved.
Citation Count: 0
Similarity-inclusive Link Prediction with Quaternions
(Scitepress, 2021) Kurt, Zuhal; Gerek, Omer Nezih; Bilge, Alper; Ozkan, Kemal; Computer Engineering
This paper proposes a Quaternion-based link prediction method, a novel representation learning method for recommendation purposes. The proposed algorithm depends on and computation with Quaternion algebra, benefiting from the expressiveness and rich representation learning capability of the Hamilton products. The proposed method depends on a link prediction approach and reveals the significant potential for performance improvement in top-N recommendation tasks. The experimental results indicate the superior performance of the approach using two quality measurements - hits rate, and coverage - on the Movielens and Hetrec datasets. Additionally, extensive experiments are conducted on three subsets of the Amazon dataset to understand the flexibility of this algorithm to incorporate different information sources and demonstrate the effectiveness of Quaternion algebra in graph-based recommendation algorithms. The proposed algorithms obtain comparatively higher performance, they are improved with similarity factors. The results show that the proposed quaternion-based algorithm can effectively deal with the deficiencies in graph-based recommender system, making it a preferable alternative among the other available methods.
Citation Count: 4
Using deep learning approaches for coloring silicone maxillofacial prostheses: A comparison of two approaches
(Wolters Kluwer Medknow Publications, 2023) Kurt, Meral; Kurt, Zuhal; Isik, Sahin; Computer Engineering
Aim: This study aimed to compare the performance of two deep learning algorithms, attention-based gated recurrent unit (GRU), and the artificial neural networks (ANNs) algorithm for coloring silicone maxillofacial prostheses. Settings and Design: This was an in vitro study. Materials and Methods: A total of 21 silicone samples in different colors were produced with four pigments (white, yellow, red, and blue). The color of the samples was measured with a spectrophotometer, then the LFNx01, aFNx01, and bFNx01 values were recorded. The relationship between the LFNx01, aFNx01, and bFNx01 values of each sample and the amount of each pigment in the compound of the same sample was used as the training dataset, entered into each algorithm, and the prediction models were obtained. While generating the prediction model for each sample, the data of the corresponding sample assigned as the target color were excluded. LFNx01, aFNx01, and bFNx01 values of each target sample were entered into the obtained models separately, and recipes indicating the ratios for mixing the four pigments were predicted. The mean absolute error (MAE) and root mean square error (RMSE) values between the original recipe used in the production of each silicone and the recipe created by both prediction models for the same silicone were calculated. Statistical Analysis Used: Data were analyzed with the Student t-test (alpha=0.05). Results: The mean RMSE values and MAE values for the ANN algorithm (0.029 & PLUSMN; 0.0152 and 0.045 & PLUSMN; 0.0235, respectively) were found significantly higher than the attention-based GRU model (0.001 & PLUSMN; 0.0005 and 0.002 & PLUSMN; 0.0008, respectively) (P < 0.001). Conclusions: Attention-based GRU model provided better performance than the ANN algorithm with respect to the MAE and RMSE values.