Bilgisayar Mühendisliği Bölümü Koleksiyonu

Bu koleksiyon için kalıcı URI

https://hdl.handle.net/20.500.14034/34

Listeleniyor 1 - 20 / 63

E-commerce product categorization using big data analytics
(2021) Usluoğlu, Sedat; Kılınç, Deniz; Bozyiğit, Fatma
E-commerce platforms need to have a well-managed online product catalog to make products easily accessible. However, the organization of catalog and categorization of products can be time-consuming due to the large volume of product data in e-commerce. In this direction, our study aims to develop an accurate categorization of product data with the adoption of big data analytics. Accordingly, various machine learning algorithms (Support Vector Machine, Naive Bayes, and Stochastic Gradient Descent) were utilized to organize online catalogs from Spark MLLib. Performed classifiers were trained and tested on product catalog data collected from a fashion retailer in Turkey, Boyner Group, which combines cutting-edge digital services with a vast network of exciting stores.
A decision support system on artificial intelligence based early diagnosis of sepsis
(2022) Kaya Aksoy, Pınar; Erdemir, Fatih; Kılınç, Deniz; Er, Orhan
Sepsis is the intense reaction of the immune system as a result of a severe infection in any part of the body and damages to organs and tissues. And this disease is commonly fatal and costly. In this study, we perform a comparative study for Sepsis prediction using machine learning algorithms from original laboratory findings. For this purpose, thirty-two different machine learning algorithms including different tructures as well as neural network classifiers are evaluated and compared. As a result of experimental studies, SVM (Cubic, Fine Gaussian), KNN (Fine, Weighted, Subspace), Trees (Weighted, Boosted, Bagged) and neural network-based classifiers have achieved a significant success rate in the diagnosis of Sepsis using the new dataset. Thus, it is concluded that it is appropriate to use machine learning algorithms to predict whether a Sepsis patient will be survived. This study has the potential to be used as a new supportive tool for doctors when predicting Sepsis.
Fraud detection on e-commerce transactions using machine learning techniques
(İzmir Bakırçay Üniversitesi, 2023) Golyeri, Murat; Celik, Sedat; Bozyigit, Fatma; Kılınç, Deniz
Fraud detection is an important aspect of e-commerce transactions as it helps to prevent fraudulent activities such as unauthorized transactions, identity theft, and account takeovers. Recently, machine learning algorithms have been widely used in the literature to detect fraud in e-commerce transactions. These algorithms work by learning patterns in the data that indicate fraudulent activity. Pattern detection involves discovering the discriminative features in the data, such as unusual transaction amounts, locations, or behaviors that are out of the normal range for a particular user, to feed the machine learning method. In this study, four basic machine learning algorithms (decision tree, logistic regression, random forest, and extreme gradient boosting) are used to detect fraud in e-commerce transactions using a newly created dataset including various features about online shopping activities on Boyner Group's e-commerce website and mobile application. The study contributes to the literature by trying different machine learning classifiers and utilizing different features that differ from current approaches in the literature.
Artificial intelligence based chatbot in e-health system
(İzmir Bakırçay Üniversitesi, 2023) Akarsu, Kamil; Er, Orhan
The healthcare sector is undergoing a digital revolution due to the rapid growth of technology, and AI technologies are becoming more commonplace in the sector. Chatbots have become useful resources for people to get advice and information about their health issues. The creation and implementation of an AI-based chatbot, integrated with an e-health system, is the main topic of this article. This paper explains the development and creation of chatbots. The chatbot's language comprehension and response capabilities are enhanced through the use of AI techniques such as machine learning and natural language processing (NLP). In addition, the chatbot's user interaction procedure and data security precautions are covered. The paper also examines how the developed chatbot can be integrated into an e-health platform and provides the results of user testing. These evaluations focus on the chatbot's ability to provide accurate and insightful answers, understand user requirements, and provide useful advice. The test results show favourable user evaluations and indicate how well the AI-based chatbot performs in providing healthcare services.
Machine learning based security analysis: Alarm generation and threat forecasting
(Ahmet Ali SÜZEN, 2020) Bozyiğit, Fatma; Türksever, Okan; Türksever, Ozan; Kılınç, Deniz
Log files keep activity records of each process performed have an important place in terms of security. Systems that provide infrastructure for applications such as network security mainly work on log management. Recently, when the security mechanisms of popular applications are examined, it has been observed that they aim to strengthen their infrastructures with machine learning (ML) methods, but in some respects, they have shortcomings. In this study, we aim to develop an alarm and security reporting system using ML methods. Our study differs from the others since it considers five separate feature (IP reputation, web reputation, malware destination access, botnet) and includes them into ML model.
A machine learning based predictive analysis use case for eSports games
(İzmir Bakırçay Üniversitesi, 2023) Tuzcu, Atakan; Ay, Emel Gizem; Uçar, Ayşegül Umay; Kılınç, Deniz
League of Legends (LoL) is a popular multiplayer online battle arena (MOBA) game that is highly recognized in the professional esports scene due to its competitive environment, strategic gameplay, and large prize pools. This study aims to predict the outcome of LoL matches and observe the impact of feature selection on model performance using machine learning classification algorithms on historical game data obtained through the official API provided by Riot Games. Detailed examinations were conducted at both team and player levels, and missing data in the dataset were addressed. A total of 1045 data were used for training team-based models, and 5232 data were used for training player-based models. Seven different machine learning models were trained and their performances were compared. Models trained on team data achieved the highest accuracy of over 98% with the AdaBoost algorithm. The top 10 features that had the most impact on the prediction outcome were identified among the 47 features in the dataset, and a new dataset was created from team data to retrain the models. After feature selection, the results showed that the accuracy of Logistic Regression increased from 89% to 98% and the accuracy of Gradient Boosting algorithm increased from 96% to 98%.
Regression based risk analysis in life insurance industry
(Ahmet Ali SÜZEN, 2020) Bozyiğit, Fatma; Şahin, Murat; Gündüz, Tolga; Işık, Cem; Kılınç, Deniz
Risk analysis is a crucial part for classifying applicants in life insurance business. Since the traditional underwriting strategies are time-consuming, recent works have focused on machine learning based methods to make the steps of underwriting more effective and strengthening the supervisory. The aim of this study is to evaluate the linear and non-linear regression-based models to determine the degree of risk. Therefore, four linear and non-linear regression algorithms are trained and evaluated on a life insurance dataset. The parameters of algorithms are optimized using Grid Search approach. The experimental results show that the non-linear regression models achieve more accurate predictions than linear regression models and the LGBM algorithm has the best performance among the all regression models with the highest R2, lowest MAE and RMSE values.
Bölgesel tabanlı evrişimli sinir ağı ile araç plaka tanıma
(Düzce Üniversitesi, 2023) Çay, Talip; Ölmez, Emre; Er, Orhan
Bu çalışmada, Bölgesel Tabanlı Evrişimli Sinir Ağları (R-CNN) ile araç plaka lokasyonu belirleme ve belirlenen lokasyon içerisinden plaka okuma işlemi gerçekleştirilmiştir. İki aşamadan oluşan çalışmanın ilk aşamasında giriş görüntüleri üzerinden plaka lokasyonları R-CNN ile belirlenirken ikinci aşamada geleneksel görüntü işleme teknikleri ile belirlenen lokasyonlardan plaka okuma işlemi gerçekleştirilmektedir. Çalışmada tasarlanan R-CNN eğitiminde veri setinde bulunan 550 adet görüntüden 450 adedi eğitimde ve 100 adedi test işleminde kullanılmıştır. R-CNN ile plaka lokasyonu bulma işleminde test seti üzerinde %95 başarı oranına ulaşılırken doğru olarak belirlenen lokasyonlardan plaka okuma işleminde %97 başarı oranına ulaşılmıştır.
Evaluating the relationship between climate change, food prices, and poverty: empirical evidence from underdeveloped countries
(Springer, 2024) Acci, Yunus; Uçar, Emine; Uçar, Murat; Acci, Reyhan Cafri
Climate change is a critical global issue with wide-ranging impacts, particularly on agriculture. This study examines how climate change influences food prices and poverty in underdeveloped countries. Rising temperatures and extreme weather events are diminishing agricultural productivity, leading to increased food prices and worsening poverty. The research involved developing a climate change index using an autoencoder model, which can learn the important features of data and translate it into a lower-dimensional representation. This index was based on variables such as carbon emission rates, annual average rainfall, forest cover, fossil fuel consumption, renewable energy use, and temperature changes. The relationship between this climate change index and food prices and poverty was analyzed using panel causality methods. Additionally, food prices from 2020 to 2030 were projected using various time series forecasting techniques to determine the most accurate predictive model. The findings indicate that while climate change does not significantly affect poverty when considering all countries as a panel, it does have a notable impact on food prices. This underscores the need for effective policy measures to address the effects of climate change on food costs. To mitigate these impacts, it is essential for policymakers to enhance agricultural resilience through sustainable practices and targeted interventions. Future research should expand the dataset and include a broader range of countries to gain a more comprehensive understanding of how climate change affects food prices and poverty.
Semantic and structural analysis of MIMIC-CXR radiography reports with NLP methods
(Gazi Univ, 2024) Uslu, Ege Erberk; Sezer, Emine; Güven, Zekeriya Anıl
Artificial intelligence that aims to imitate human decision-making processes, using human knowledge as a foundation, is a critical research area with various practical applications in different disciplines. In the health domain, machine learning and image processing techniques are increasingly being used to assist in diagnosing diseases. Many healthcare reports, such as epicrisis summaries prepared by clinical experts, contain crucial and valuable information. In addition to information extraction from healthcare reports, applications such as automatic healthcare report generation are among the natural language processing research areas based on this knowledge and experience. The primary goals are to reduce the workload of clinical experts, minimize the likelihood of errors, and save time to speed up the diagnosis process. The MIMIC-CXR dataset is a huge dataset consisting of chest radiographs and reports prepared by radiology experts related to these images. Before developing a natural language processingbased model, preprocessing steps were applied to the dataset, and the results of syntactic and semantic analyses performed on unstructured report datasets are presented. The results show that most examined words and phrases exhibit minimal semantic inference disparities. The generic named entity recognition method demonstrates comparatively lower effectiveness than the ngram technique in extracting word frequencies. However, named entity recognition facilitated the identification of medical entities within the dataset. This study is expected to provide insights for developing language models, particularly for developing a natural language processing model on the MIMIC-CXR dataset.
A decision support system for machine learning-based determination of zinc deficiency: A study in adolescent patients
(Brieflands, 2024) Orbatu, Dilek; Bulgan, Zeynep Izem Peker; Ölmez, Emre; Er, Orhan
Background: Over the past three years, zinc deficiency among adolescents has varied based on region and access to healthcare. Globally, zinc deficiency affects approximately 2 billion people, leading to serious issues such as immune problems and growth delays, particularly in developing countries. In the U.S., around 10% of adolescents experienced zinc deficiency in 2021, with a higher prevalence among teenage girls. In Europe, deficiency rates are generally low but can be significant in Eastern Europe and Central Asia. In Asia, particularly in rural and low-income areas, deficiency rates range from 20- 30%. In Turkey, the prevalence is high due to poor nutrition. Objectives: This study aimed to develop a machine learning-based decision support system to determine zinc deficiency in children and adolescents aged ID- 18 years. Methods: This machine learning-based study was conducted with 370 adolescents aged 10-18 years to assess their zinc deficiency. The dataset consists of 8 feature vectors and an output vector. The machine learning methods used in the analysis include logistic regression, naive bayes, decision tree (CART), K-nearest neighbors (K-NN), support vector machine (SVM), gradient boosting classifier, AdaBoost classifier; bagging classifier; random forest classifier; multilayer perceptron (MLP) classifier; and XGBoost (XGB) classifier. Evaluation metrics such as accuracy, precision, recall, and Fl score were used to assess the performance of these methods. Including specific values for these metrics, such as SVM achieved 94.6% accuracy, would allow readers to quicldy compare the effectiveness of the models. Different metrics serve various purposes: Accuracy provides an overall view of performance, precision and recall highlight specific aspects, and the Fl score balances precision and recall. Results: The mean age of the patients in the dataset was 13.79 +/- 1.18 years. Of the children, 6432% (n = 238) were female and 35.68% (n =B2) were male. It was found that 62.7% (n = 232) of the children had low zinc levels, while 373% (n = ox ) did not require zinc supplementation. Thirteen different machine learning methods were applied to a 70% training and 30% testing set. As a result, the SVM method provided the most successful outcome with 94.6% accuracy. Implementing the SVM-based system in pediatric clinics could improve efficiency and patient care by automatically detecting high-risk zinc deficiency patients based on lab results, providing early intervention alerts for faster treatment, and improving health outcomes. Highlighting these practical applications could increase the study's appeal to healthcare professionals by demonstrating its real-world benefits. Providing detailed information on these applications would enhance the study's clarity and practical value, making it more valuable for researchers and healthcare providers interested in Al tools for adolescent health. Conclusions: This study concluded that machine learning methods can effectively determine zinc deficiency in children. The SVM method demonstrated superior classification performance compared to the other methods. An SVM-based decision support system could be integrated into pediatric outpatient clinics to enhance diagnostic accuracy and patient care.
Real-time chord identification application: Enabling lifelong music education through seamless integration of audio processing and machine learning
(Nilgun SAZAK, 2024) Özbaltan, Nihan
Lifelong music education is critical need for all with a particular focus on adult learners and seniors. One of the difficulties in music education is identifying chords accurately. This is a preliminary study to develop a chord identification application using Artificial Intelligence (AI) technologies. I seek to answer the key research question of how audio processing algorithms and deep learning models can be used to provide real-time, accurate and user-friendly chord recognition that meets the diverse needs of adult learners and senior citizens. Our overall goal is to create an application that not only assists with chord identification, but also fosters a lifelong love of music and learning. My methodology is based on the principles of adult and senior education initiatives and includes the following key steps: using ready-made datasets for audio processing and feature extraction, transforming waveforms into mel spectrograms, and preparing and extending the datasets where necessary. I then train and optimise deep learning models, such as various convolutional neural network (CNN) architectures, to achieve high accuracy in chord recognition. By using advanced technologies and adhering to the principles of lifelong learning, our research aims to enhance the musical journey of individuals throughout their lives, contributing to both personal enrichment and cognitive well-being. © 2024 The Author(s).
The impact of artificial intelligence on healthcare industry: Volume 1: Non-clinical applications
(CRC Press, 2024) Berktaş, Mustafa; Hızıroğlu, Abdulkadir; Erbaycu, Ahmet Emin; Er, Orhan; Kahyaoğlu, Sezer Bozkuş
Healthcare and medical science are inherently dependent on technological advances and innovations for improved care. In recent times we have witnessed a new drive in implementing these advances and innovations through the use of Artificial Intelligence, in both clinical and non-clinical areas. The set of 2 volumes aims to make available the latest research and applications to all, and to present the current state of clinical and non-clinical applications in the health sector and areas open to development, as well as to provide recommendations to policymakers. This volume covers non-clinical applications. The chapters covered in this book have been written by professionals who are experts in the healthcare sector and have academic experience. © 2025 Mustafa Berktas, Abdulkadir Hiziroglu, Ahmet Emin Erbaycu, Orhan Er and Sezer Bozkus Kahyaoglu.
Development of context-sensitive formulas to obtain constant luminance perception
(Institute of Electrical and Electronics Engineers Inc., 2024) Akleman, Ergun; Akgün, Bekir Tevfik; Alpkoçak, Adil
In this article, we present a method to develop context-sensitive luminance correction formulas to produce constant luminance perception for a foreground object. Our formulas make the foreground object slightly transparent to mix with the blurred version of the background. This mix can be used to quickly produce any desired illusion of luminance in foreground objects based on the luminance of the background. The transparency formula has only one variable; the relative size of the foreground object. We identified the general structure of the transparency formula as a function of the relative size of the foreground object. We develop a method to intuitively control the coefficients of the transparency formula. We have implemented an interactive Web-based program in Shadertoy to identify coefficients. Using this program, we determined the coefficients of the transparency formula. We also identified a simpler affine formula, which requires only two coefficients. We made our program publicly available so that anyone can access and improve it. In this article, we also explain how to intuitively change the polynomial part of the formula. Using our explanation, users change the polynomial part of the formula to obtain their own perceptively constant luminance. This can be used as a crowd-sourcing experiment to further improve the formula. © 2024 IEEE.
NLP-powered healthcare insights: A comparative analysis for multi-labeling classification with MIMIC-CXR dataset
(Institute of Electrical and Electronics Engineers Inc., 2024) Erberk Uslu, Ege; Sezer, Emine; Anıl Güven, Zekeriya
The digitization of the healthcare industry has led to a growing number of applications that use machine learning and image processing techniques to improve the diagnostic process. These applications utilize a variety of medical data, including laboratory results, clinical findings, MRI scans, tomographic images, and radiological images. In addition, free-text healthcare documentation, such as well-structured discharge summaries, contains valuable information. Natural Language Processing encompasses the development of automated systems for generating health reports. This process involves using domain-specific knowledge and prior knowledge to extract relevant information from medical records. This article investigates the use of natural language processing techniques for chest X-ray classification. A total of 14 distinct impressions derived from chest radiography findings from the MIMIC-CXR dataset were used in a multi-label classification procedure. Six distinct language models derived from the BERT language model, along with three distinct classification algorithms, were employed to evaluate the effectiveness of the models and the dataset for multi-label categorization. The experimental results showed a successful prediction rate of 80.47% for 14 distinct impressions within the dataset. © 2024 The Authors.
Applying machine learninh to audit data: Enhancing fraud detection, risk assessment and audit efficiency
(Taylor and Francis Ltd., 2024) Özbaltan, Nihan
Machine learning (ML) is used globally as a tool for predictive analysis. Within auditing, the use of audit data helps to uncover fraud indicators, identify risk areas and implement predictive models for continuous audit monitoring. Researchers are using various machine learning methods to analyze large and complex audit data to facilitate prediction. In this study, an online UCI dataset of 776 lines and 27 features is used. Out of these 27 features, 13 are eliminated due to their low impact on the target dataset or due to the ‘important feature selection’ algorithm. In this analysis, I used supervised learning methods, namely K-Nearest Neighbors, Logistic Regression, Random Forest, Support Vector Machine, Decision Tree, Linear Discriminant Analysis, Gaussian Naive Bayes, Extra Tree Algorithm, Gradient Boosting Algorithm, Ada Boosting Algorithm and XGBoost Algorithms. The experimental results highlight the power of eight neighbor KNN and evaluate its effectiveness, sensitivity, precision, accuracy and F1 score in comparison with other methods such as Naive Bayes, SVM (Linear Kernel), Decision Tree Classifier and Random Forest Classifier. © Copyright 2024 Taylor & Francis–All rights reserved.
Classification of radiographic and non-radiographic axial spondylarthritis in pelvic radiography using deep convolution neural network models
(Springer, 2025) Kahveci, Abdulvahap; Alcan, Veysel; Uçar, Murat; Gümüştepe, Alper; Bilgin, Esra; Sunar, İsmihan; Ataman, Şebnem
Discriminating radiographic axial spondyloarthritis (r-axSpA) from nonradiographic axial spondyloarthritis (nr-axSpA) using pelvic radiographs is challenging, especially for inexperienced clinicians. This study aims to perform deep convolution neuronal network (CNN) models to aid in this diagnostic challenge by using their radiographs. Six-hundred sacroiliac joint exams (300 pelvic radiographs) of patients from axSpA cohort were enrolled (screened between Jan 2010 and Jan 2020). All radiographs were examined and graded by a blinded expert rheumatologist. Four CNN models (VGG16, ResNet, DenseNet, and MobileNet) were proposed by combining them with the YOLOv7 object detection algorithm to mark the sacroiliac joints. The classification results of CNNs were evaluated by performance metrics [accuracy, AUROC (area under the receiver operating characteristic curve)]. The VGG16 model with the YOLOv7 algorithm yielded the best performance [accuracy of 83.8% (95% CI; 73.3–92.9%)]. The accuracy values of other models were 70.7% (58.3–82.7%), 77.1% (65.1–87.3%), and 71.8% (59.0-83.1%) for ResNet, DenseNet, and MobileNet, respectively. In the ROC analysis, the AUC value of the VGG16 model (AUC = 0.882) was higher than other CNNs (AUCs = 0.836, 0.808, and 0.787; DenseNet, ResNet, and MobileNet, respectively). This paper revealed deep learning architectures were able to differentiate r-axSpA from nr-axSpA on pelvic radiographs. Hereby, these models might be used as a clinical decision support system in clinical practice. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024.
Comprehensive analysis study of techniques in different domains for Turkish music genre classification task
(Springer Science and Business Media Deutschland GmbH, 2025) Güven, Zekeriya Anıl
Nowadays, models or algorithms are used in the analysis process as the amount of data increases. Depending on the sectors, techniques in domains such as NLP, image processing, and voice analysis can be used. In this study, analyses were applied in these domains to classify music genres on the Turkish music dataset and these domains were compared. To perform the first analysis, the voice characteristic features of the songs were extracted and the success of machine learning (ML) algorithms and the CNN model were analyzed. For the next analysis, spectrograms of the songs were extracted and Keras application models were trained with transfer learning. During these analyses, audio segmentation and feature reduction techniques were also performed on the songs to analyze them. The last analysis applied textual analysis with song lyrics to the NLP domain. After preprocessing, the vector representations of these lyrics were obtained and the success of ML algorithms and the CNN model was measured. At the same time, large language models were fine-tuned and the success of these models was analyzed. As a result of all analyses, it has been shown that the ML method with the application of audio segmentation and feature reduction for voice analysis is the most successful. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2024.
Anomaly detection for gear manufacturing in the automotive sector using rare itemset mining
(2022) BİRANT, Derya; TAŞER, Pelin YILDIRIM; Akdaş, Devrim Naz
Downtimes in manufacturing significantly influence productivity, and their analysis is necessary for successful and flexible production. Although some classification and regression studies have been performed on the machine downtime in the manufacturing area, the rare itemset mining (RIM) technique has never been implemented in the existing downtime studies until now. Besides, anomaly detection for gear manufacturing downtime in the automotive sector using RIM is yet to be explored. To bridge this gap, this study proposes the application of the RIM method for detecting anomalies in gear manufacturing downtime of earth moving machinery for the first time. In this study, the Rare Pattern Growth (RP-Growth) algorithm was executed on a real-world dataset consisting of downtimes in gear manufacturing of earth moving machinery to discover rare itemsets that indicate anomalies in downtimes. In the experiments, the rare itemsets (anomalies) in the downtime data were detected using different minimum support (minsup) and minimum rare support (minraresup) threshold values. The obtained results were also evaluated in terms of the number of itemsets, execution time, and maximum memory usage. The experimental results show that the proposed approach, called Anomaly Detection with Rare Itemset Mining (ADRIM), is an effective method for detecting anomalies in machine downtimes and can be successfully used in the manufacturing area, especially in the automotive sector.
Machine learning models for early prediction of mortality risk in patients with burns: A single center experience
(Churchill Livingstone, 2024) Çinar M.A.; Ölmez Emre; Erkiliç A.; Bayramlar K.; Er, Orhan
Mortality rate is considered as the most important outcome measure for assessing the severity of burn injury. A scale or model that accurately predicts burn mortality can be useful to determine the clinical course of burn injuries, discuss treatment options and rehabilitation with patients and their families, and evaluate novel, innovative interventions for the injuries. This study aimed to use machine learning models to predict the mortality risk of patients with burns after their first admission to the center and to compare the performances of these models. Overall, 1064 patients hospitalized in burn intensive care and burn service units between 2016 and 2022 were included in the study. In total, 40 parameters, including demographic characteristics and biochemical parameters of all patients, were analyzed in the study. Furthermore, the dataset was randomly divided into two clusters with 70% of the data used for artificial neural networks (ANNs) training and 30% for model success testing. The ANN model proposed in this study showed high success across all machine learning methods tried in different variants, with an accuracy of 95.92% in the test set. Machine learning models can be used to predict the mortality risk of patients with burns. This study may help validate the use of machine learning models for applications in clinical practice. Conducting multicenter studies will further contribute to the literature. © 2023 British Association of Plastic, Reconstructive and Aesthetic Surgeons

Güncel Gönderiler