Bilgisayar ve Bilişim Fakültesi Koleksiyonu

Permanent URI for this collectionhttps://hdl.handle.net/20.500.13091/10834

Browse

Now showing 1 - 20 of 62

A 3d U-Net Based on Early Fusion Model: Improvement, Comparative Analysis With State-Of Models and Fine-Tuning
(Konya Teknik Univ, 2024) Kayhan, Beyza; Uymaz, Sait Ali
Multi-organ segmentation is the process of identifying and separating multiple organs in medical images. This segmentation allows for the detection of structural abnormalities by examining the morphological structure of organs. Carrying out the process quickly and precisely has become an important issue in today's conditions. In recent years, researchers have used various technologies for the automatic segmentation of multiple organs. In this study, improvements were made to increase the multi-organ segmentation performance of the 3D U-Net based fusion model combining HSV and grayscale color spaces and compared with state-of-the-art models. Training and testing were performed on the MICCAI 2015 dataset published at Vanderbilt University, which contains 3D abdominal CT images in NIfTI format. The model's performance was evaluated using the Dice similarity coefficient. In the tests, the liver organ showed the highest Dice score. Considering the average Dice score of all organs, and comparing it with other models, it has been observed that the fusion approach model yields promising results.
Citation - WoS: 29
Citation - Scopus: 54
Alexnet Architecture Variations With Transfer Learning for Classification of Wound Images
(Elsevier B.V., 2023) Eldem, H.; Ülker, E.; Işıklı, O.Y.
In medical world, wound care and follow-up is one of the issues that are gaining importance to work on day by day. Accurate and early recognition of wounds can reduce treatment costs. In the field of computer vision, deep learning architectures have received great attention recently. The achievements of existing pre-trained architectures for describing (classifying) data belonging to many image sets in the real world are primarily addressed. However, to increase the success of these architectures in a certain area, some improvements and enhancements can be made on the architecture. In this paper, the classification of pressure and diabetic wound images was performed with high accuracy. The six different new AlexNet architecture variations (3Conv_Softmax, 3Conv_SVM, 4Conv_Softmax, 4Conv_SVM, 6Conv_Softmax, 6Conv_SVM) were created with a different number of implementations of Convolution, Pooling, and Rectified Linear Activation (ReLU) layers. Classification performances of the proposed models are investigated by using Softmax classifier and SVM classifier separately. A new original Wound Image Database are created for performance measures. According to the experimental results obtained for the Database, the model with 6 Convolution layers (6Conv_SVM) was the most successful method among the proposed methods with 98.85% accuracy, 98.86% sensitivity, and 99.42% specificity. The 6Conv_SVM model was also tested on diabetic and pressure wound images in the public medetec dataset, and 95.33% accuracy, 95.33% sensitivity, and 97.66% specificity values were obtained. The proposed method provides high performance compared to the pre-trained AlexNet architecture and other state-of-the-art models in the literature. The results showed that the proposed 6Conv_SVM architecture can be used by the relevant departments in the medical world with good performance in medical tasks such as examining and classifying wound images and following up the wound process. © 2023 Karabuk University
Citation - WoS: 3
Citation - Scopus: 6
Analysis of Machine Learning Classification Approaches for Predicting Students' Programming Aptitude
(MDPI, 2023) Çetinkaya, Ali; Baykan, Ömer Kaan; Kırgız, Havva
With the increasing prevalence and significance of computer programming, a crucial challenge that lies ahead of teachers and parents is to identify students adept at computer programming and direct them to relevant programming fields. As most studies on students' coding abilities focus on elementary, high school, and university students in developed countries, we aimed to determine the coding abilities of middle school students in Turkey. We first administered a three-part spatial test to 600 secondary school students, of whom 400 completed the survey and the 20-level Classic Maze course on Code.org. We then employed four machine learning (ML) algorithms, namely, support vector machine (SVM), decision tree, k-nearest neighbor, and quadratic discriminant to classify the coding abilities of these students using spatial test and Code.org platform data. SVM yielded the most accurate results and can thus be considered a suitable ML technique to determine the coding abilities of participants. This article promotes quality education and coding skills for workforce development and sustainable industrialization, aligned with the United Nations Sustainable Development Goals.
Citation - Scopus: 1
Automatic Sleep Stage Classification for the Obstructive Sleep Apnea
(Trans Tech Publications Ltd, 2023) Özsen, Seral; Koca, Yasin; Tezel, Gülay Tezel; Solak, Fatma Zehra; Vatansev, Hulya; Kucukturk, Serkan
Automatic sleep scoring systems have been much more attention in the last decades. Whereas a wide variety of studies have been used in this subject area, the accuracies are still under acceptable limits to apply these methods to real-life data. One can find many high-accuracy studies in literature using a standard database but when it comes to using real data reaching such high performance is not straightforward. In this study, five distinct datasets were prepared using 124 persons including 93 unhealthy and 31 healthy persons. These datasets consist of time-, nonlinear-, welch-, discrete wavelet transform- and Hilbert-Huang transform features. By applying k-NN, Decision Trees, ANN, SVM, and Bagged Tree classifiers to these feature sets in various manners by using feature-selection highest classification accuracy was searched. The maximum classification accuracy was detected in the case of the Bagged Tree classifier as 95.06% with the use of 14 features among a total of 136 features. This accuracy is relatively high compared with the literature for a real-data application.
Citation - WoS: 30
Citation - Scopus: 36
Binary Aquila Optimizer for 0-1 Knapsack Problems
(Pergamon-Elsevier Science Ltd, 2023) Baş, Emine
The optimization process entails determining the best values for various system characteristics in order to finish the system design at the lowest possible cost. In general, real-world applications and issues in artificial intelligence and machine learning are discrete, unconstrained, or discrete. Optimization approaches have a high success rate in tackling such situations. As a result, several sophisticated heuristic algorithms based on swarm intelligence have been presented in recent years. Various academics in the literature have worked on such algorithms and have effectively addressed many difficulties. Aquila Optimizer (AO) is one such algorithm. Aquila Optimizer (AO) is a recently suggested heuristic algorithm. It is a novel population-based optimization strategy. It was made by mimicking the natural behavior of the Aquila. It was created by imitating the behavior of the Aquila in nature in the process of catching its prey. The AO algorithm is an algorithm developed to solve continuous optimization problems in their original form. In this study, the AO structure has been updated again to solve binary optimization problems. Problems encountered in the real world do not always have continuous values. It exists in problems with discrete values. Therefore, algorithms that solve continuous problems need to be restructured to solve discrete optimization problems as well. Binary optimization problems constitute a subgroup of discrete optimization problems. In this study, a new algorithm is proposed for binary optimization problems (BAO). The most successful BAO-T algorithm was created by testing the success of BAO in eight different transfer functions. Transfer functions play an active role in converting the continuous search space to the binary search space. BAO has also been developed by adding candidate solution step crossover and mutation methods (BAO-CM). The success of the proposed BAO-T and BAO-CM algorithms has been tested on the knapsack problem, which is widely selected in binary optimization problems in the literature. Knapsack problem examples are divided into three different benchmark groups in this study. A total of sixty-three low, medium, and large scale knapsack problems were determined as test datasets. The performances of BAO-T and BAO-CM algorithms were examined in detail and the results were clearly shown with graphics. In addition, the results of BAO-T and BAO-CM algorithms have been compared with the new heuristic algorithms proposed in the literature in recent years, and their success has been proven. According to the results, BAO-CM performed better than BAO-T and can be suggested as an alternative algorithm for solving binary optimization problems.
Citation - WoS: 10
Citation - Scopus: 11
A Binary Sparrow Search Algorithm for Feature Selection on Classification of X-Ray Security Images
(Elsevier Ltd, 2024) Babalik, A.; Babadag, A.
In today's world, especially in public places, strict security measures are being implemented. Among these measures, the most common is the inspection of the contents of people's belongings, such as purses, knapsacks, and suitcases, through X-ray imaging to detect prohibited items. However, this process is typically performed manually by security personnel. It is an exhausting task that demands continuous attention and concentration, making it prone to errors. Additionally, the detection and classification of overlapping and occluded objects can be challenging. Therefore, automating this process can be highly beneficial for reducing errors and improving the overall efficiency. In this study, a framework consisting of three fundamental phases for the classification of prohibited objects was proposed. In the first phase, a deep neural network was trained using X-ray images to extract features. In the subsequent phase, features that best represent the object were selected. Feature selection helps eliminate redundant features, leading to the efficient use of memory, reduced computational costs, and improved classification accuracy owing to a decrease in the number of features. In the final phase, classification was performed using the selected features. In the first stage, a convolutional neural network model was utilized for feature extraction. In the second stage, the Sparrow Search Algorithm was binarized and proposed as the binISSA for feature selection. Feature selection was implemented using the proposed binISSA. In the final stage, classification was performed using the K-Nearest Neighbors (KNN) and Support Vector Machine (SVM) algorithms. The performances of the convolutional neural network and the proposed framework were compared. In addition, the performance of the proposed framework was compared with that of other state-of-the-art meta-heuristic algorithms. The proposed method increased the classification accuracy of the network from 0.9702 to 0.9763 using both the KNN and SVM (linear kernel) classifiers. The total number of features extracted using the deep neural network was 512. With the application of the proposed binISSA, average number of features were reduced to 25.33 using the KNN classifier and 32.70 using the SVM classifier. The results indicate a notable reduction in the extracted features from the convolutional neural network and an improvement in the classification accuracy. © 2024 Elsevier B.V.
Citation - WoS: 7
Citation - Scopus: 9
Bindmo: a New Binary Dwarf Mongoose Optimization Algorithm on Based Z-Shaped, U-Shaped, and Taper-Shaped Transfer Functions for Cec-2017 Benchmarks
(Springer Science and Business Media Deutschland GmbH, 2024) Baş, Emine
Intelligent swarm optimization algorithms have become increasingly common due to their success in solving real-world problems. Dwarf Mongoose Optimization (DMO) algorithm is a newly proposed intelligent swarm optimization algorithm in recent years. It was developed for continuous optimization problem solutions in its original paper. But real-world problems are not always problems that take continuously variable values. Real-world problems are often problems with discrete variables. Therefore, heuristic algorithms proposed for continuous optimization problems need to be updated to solve discrete optimization problems. In this study, DMO has been updated for binary optimization problems and the Binary DMO (BinDMO) algorithm has been proposed. In binary optimization, the search space consists of binary variable values. Transfer functions are often used in the conversion of continuous variable values to binary variable values. In this study, twelve different transfer functions were used (four Z-shaped, four U-shaped, and four Taper-shaped). Thus, twelve different BinDMO variations were obtained (BinDMO1, BinDMO2, …, BinDMO12). The achievements of BinDMO variations were tested on thirteen different unimodal and multimodal classical benchmark functions. The effectiveness of population sizes on the effectiveness of BinDMO was also investigated. When the results were examined, it was determined that the most successful BinDMO variation was BinDMO1 (with Z1-shaped transfer function). The most successful BinDMO variation was compared with three different binary heuristic algorithms selected from the literature (SO, PDO, and AFT) on CEC-2017 benchmark functions. According to the average results, BinDMO was the most successful binary heuristic algorithm. This has proven that BinDMO can be chosen as an alternative algorithm for binary optimization problems. © The Author(s) 2024.
Citation - Scopus: 15
Chaos Theory in Metaheuristics
(Elsevier, 2023) Türkoğlu, B.; Uymaz, S.A.; Kaya, E.
Metaheuristic optimization is the technique of finding the most suitable solution among the possible solutions for a particular problem. We encounter many problems in the real world, such as timetabling, path planning, packing, traveling salesman, trajectory optimization, and engineering design problems. The two main problems faced by all metaheuristic algorithms are being stuck in local optima and early convergence. To overcome these problems and achieve better performance, chaos theory is included in the metaheuristic optimization. The chaotic maps are employed to balance the exploration and exploitation efficiently and improve the performance of algorithms in terms of both local optima avoidance and convergence speed. The literature shows that chaotic maps can significantly boost the performance of metaheuristic optimization algorithms. In this chapter, chaos theory and chaotic maps are briefly explained. The use of chaotic maps in metaheuristic is presented, and an enhanced version of GSA with chaotic maps is shown as an application. © 2023 Elsevier Inc. All rights reserved.
Citation - WoS: 11
Citation - Scopus: 12
Chaotic Artificial Algae Algorithm for Solving Global Optimization With Real-World Space Trajectory Design Problems
(Springer Heidelberg, 2025) Turkoğlu, Bahaeddin; Uymaz, Sait Ali; Kaya, Ersin
The artificial algae algorithm (AAA) is a recently introduced metaheuristic algorithm inspired by the behavior and characteristics of microalgae. Like other metaheuristic algorithms, AAA faces challenges such as local optima and premature convergence. Various strategies to address these issues and enhance the performance of the algorithm have been proposed in the literature. These include levy flight, local search, variable search, intelligent search, multi-agent systems, and quantum behaviors. This paper introduces chaos theory as a strategy to improve AAA's performance. Chaotic maps are utilized to effectively balance exploration and exploitation, prevent premature convergence, and avoid local minima. Ten popular chaotic maps are employed to enhance AAA's performance, resulting in the chaotic artificial algae algorithm (CAAA). CAAA's performance is evaluated on thirty benchmark test functions, including unimodal, multimodal, and fixed dimension problems. The algorithm is also tested on three classical engineering problems and eight space trajectory design problems at the European Space Agency. A statistical analysis using the Friedman and Wilcoxon tests confirms that CAA demonstrates successful performance in optimization problems.
Citation - WoS: 17
Citation - Scopus: 15
Chaotic Golden Ratio Guided Local Search for Big Data Optimization
(Elsevier - Division Reed Elsevier India Pvt Ltd, 2023) Koçer, Havva Gül; Türkoğlu, Bahaeddin; Uymaz, Sait Ali
Biological systems where order arises from disorder inspires for many metaheuristic optimization techniques. Self-organization and evolution are the common behaviour of chaos and optimization algorithms. Chaos can be defined as an ordered state of disorder that is hypersensitive to initial conditions. Therefore, chaos can help create order out of disorder. In the scope of this work, Golden Ratio Guided Local Search method was improved with inspiration by chaos and named as Chaotic Golden Ratio Guided Local Search (CGRGLS). Chaos is used as a random number generator in the proposed method. The coefficient in the equation for determining adaptive step size was derived from the Singer Chaotic Map. Performance evaluation of the proposed method was done by using CGRGLS in the local search part of MLSHADE-SPA algorithm. The experimental studies carried out with the electroencephalographic signal decomposition based optimization problems, named as Big Data optimization problem (Big-Opt), introduced at the Congress on Evolutionary Computing Big Data Competition (CEC'2015). Experimental results have shown that the local search method developed using chaotic maps has an effect that increases the performance of the algorithm.& COPY; 2023 Karabuk University. Publishing services by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Classification of Knee Osteoarthritis Severity by Transfer Learning From X-Ray Images
(2024) Solak, Fatma Zehra
Knee Osteoarthritis (KOA) is the most common type of arthritis and its severity is assessed with the Kellgren-Lawrence (KL) grading system based on evidence from both knee bones. Recent advancements point to an era where computer-assisted methods enhance KOA diagnostic efficiency. This study implemented binary and multiple classification processes based on X-ray images and deep learning algorithms for computer-aided KOA severity diagnosis. Pre-processing involved extracting the region of interest and contrast enhancement with CLAHE on the X-ray images from the included dataset. Using this dataset, 2, 3, 4, and 5 class classification processes were conducted with ResNet-50, Xception, VGG16, EfficientNetb0, and DenseNet201 transfer learning models. Each model was assessed with “rmsprop,” “sgdm,” and “adam” optimization algorithms. Study findings reveal that, the DenseNet201-rmsprop model achieved 87.7% accuracy, 87.2% F1-Score, and a 0.75 Cohen’s kappa value for 2-class classification. For 3-class classification, it achieved 85.6% accuracy, 82.4% F1-Score, and a 0.71 Cohen’s kappa value. For 4-class classification, the DenseNet201-rmsprop model provided 81.5% accuracy, 77.1% F1-Score, and a Cohen’s kappa value of 0.67. In the 5-class classification, the highest success was with the Xception-rmsprop model, with 67.8% accuracy, 68.8% F1-Score, and a 0.55 Cohen’s kappa value. The evaluation with varying class numbers and different transfer learning models highlights the proposed approach’s effectiveness. Results of the study underscore the study’s uniqueness and success in demonstrating how varying the number of classes, employing different transfer learning models and optimizers can provide clearer insights into KOA severity evaluation.
Citation - WoS: 3
Citation - Scopus: 3
Classification of Physiological Disorders in Apples Using Deep Convolutional Neural Network Under Different Lighting Conditions
(Springer, 2023) Büyükarıkan, Birkan; Ülker, Erkan
Non-destructive testing of apple fruit, an important product in the world fresh fruit trade, according to physiological disorders, can be done with a computer vision system. However, in the vision system, images may be affected by the brightness values created by different lighting conditions. For this reason, it is a necessity to use algorithms that accurately and quickly detect physiological disorders. By using a convolutional neural network (CNN), an algorithm that enables easy extraction of features from images, determining physiological disorders becomes easier. This study aims to classify the images of apples with physiological disorders obtained under different lighting conditions with CNN models. This study created a dataset (images of different light colors, angles, and distances) with some physiological disorder images. A 5-fold cross-validation method was applied to improve the generalization ability of the models, and CNN models were trained end-to-end. In addition, the Friedman hypothesis test and post-hoc Nemenyi test were performed to compare the evaluation indicators of different CNN models. The average accuracy, precision, recall, and F1-score of the Xception model were 0.996, 0.994, 0.996, and 0.998, respectively. The classification accuracy of this model is followed by the ResNet101, MobileNet, ResNet152, ResNet18, ResNet34, ResNet50, EfficientNetB0, AlexNet, VGG16, and VGG19. Finally, Xception performed well, according to Friedman/Nemenyi test results.
Çok Amaçlı Dağınık Arama Algoritmasının Zdt-dtlz Test Problemleri Üzerinde Uygulanması
(2024) Haber, Zeynep; Uğuz, Harun
Dağınık arama algoritması, tek amaçlı optimizasyon problemlerinin çözümünde sıkça kullanılan bir yöntemdir. Ancak, çok amaçlı problemlerle başa çıkmak oldukça zorlu bir süreçtir. Bu makale, çok amaçlı optimizasyon problemleriyle başa çıkabilmek için \"Dağınık Arama Algoritması\" (DA) olarak adlandırılan yöntemin genişletilmesine yönelik bir öneri sunmaktadır. Önerilen yaklaşım, DA algoritmasına çok amaçlı optimizasyon algoritması olan Baskın Olmayan Sıralama Genetik Algoritması II (NSGA-II) yöntemindeki Yoğunluk Mesafesi (CD) ve Hızlı Bastırılmamış Sıralama kavramlarını ekleyerek hibrit çok amaçlı optimizasyon algoritması önermektedir. Bu önerilen algoritma, ZDT ve DTLZ test problemleri kullanılarak değerlendirilmiştir. Yapılan deneysel sonuçlar, önerilen Çok Amaçlı Dağınık Arama(ÇADA) algoritmasının 19 farklı çok amaçlı optimizasyon yöntemi ile karşılaştırıldığında, ZDT problemi için 2.40 IGD ortalama ile birinci sırada, DTLZ probleminde ise 0.0035 IGD ortalama değeri ile altıncı sırada yer aldığını göstermektedir. Bu sonuçlar, önerilen algoritmanın karşılaştırılabilir düzeyde başarılı bir performansa sahip olduğunu ortaya koymaktadır.
Comparison of Textual Data Augmentation Methods on Sst-2 Dataset
(Springer Science and Business Media Deutschland GmbH, 2024) Çataltaş, M.; Baykan, N.A.; Cicekli, I.
Since the arrival of advanced deep learning models, more successful techniques have been proposed, significantly enhancing the performance of nearly all natural language processing tasks. While these deep learning models achieve the best results, large datasets are needed to get these results. However, data collection in large amounts is a challenging task and cannot be done successfully for every task. Therefore, data augmentation might be required to satisfy the need for large datasets by generating synthetic data samples using original data samples. This study aims to give an idea to those who will work in this field by comparing the successes of using a large dataset as a whole and data augmentation in smaller pieces at different rates. For this aim, this study presents a comparison of three textual data augmentation techniques, examining their efficacy based on the augmentation mechanism. Through empirical evaluations on the Stanford Sentiment Treebank dataset, the sampling-based method LAMBADA showed superior performance in low-data regime scenarios and moreover showcased better results than other methods when the augmentation ratio is increased, offering significant improvements in model robustness and accuracy. These findings offer insights for researchers on augmentation strategies, thereby enhancing generalization in future works. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2024.
Citation - WoS: 5
Citation - Scopus: 7
Convolutional Neural Network-Based Apple Images Classification and Image Quality Measurement by Light Colors Using the Color-Balancing Approach
(Springer, 2023) Büyükarıkan, Birkan; Ülker, Erkan
The appearance of an object is affected by the color and quality of the light on the surface and the location of the lighting source. Color-balancing methods can solve the problems caused by light changes. Color-balancing models increase the visibility of the image by changing color and clarity. The study aims to examine the images of physiological disorders in apples' classification performances of images in different light colors with color-balancing models with pre-trained CNN models. Physiological disorders were classified with 0.949 accuracies in the ResNet50V2 model and sharpness data set in the green light color. With the proposed approaches, there was an increase in performance compared to the original data set. The best success in all light colors is in the sharpness data set type. In addition, the quality of the images was measured using MSE, PSNR, and SSIM. PSNR increased in the warm and cold white sharpness data set type and green light CLAHE data set type. Finally, experimental studies have shown that color balancing significantly affects classification success.
Coronavirüs Sürü Bağışıklığı Algoritması ile Otsu Tabanlı Optimal Çok Düzeyli Görüntü Eşiği
(2023) Koç, İsmail
Eşik seçimi, görüntü bölütlemede önemli bir rol oynamaktadır. Eşik seçimiyle ilgili en faydalı yöntemler olarak minimum hata yöntemi, iteratif yöntem, entropi yöntemi ve Otsu yöntemi bilinmektedir. Bu çalışmada eşikleme yöntemi olarak Otsu tekniği kullanılmaktadır. Eşik sayısının (K) artmasına bağlı olarak problemin karmaşıklık düzeyi üstel olarak artacağı için matematiksel yöntemler yerine sürü zekâsı algoritması kullanılması daha uygun görülmektedir. Bundan dolayı, bu çalışmada sürü zekâsı algoritması olarak da son yıllarda literatüre kazandırılmış olan Coronavirüs sürü bağışıklığı algoritması (CHIO) kullanılmaktadır. Deneysel çalışmalarda test verisi olarak altı farklı görüntü kullanılmaktadır. K değeri bu çalışmada 2, 3, 4 ve 5 olarak belirlenmektedir. Bu veri seti kullanılarak CHIO algoritması ile literatürde yer alan diferansiyel evrim (differential evolution: DE), gri kurt ( gray wolf optimizer: GWO), parçacık sürü (particle swarm optimization: PSO) algoritmaları gibi başarılı algoritmalarla eşit koşullarda kıyaslanmaktadır. Elde edilen sonuçlara göre, CHIO algoritması kullanılarak 6 test verisi üzerinde yapılan çalışmalarda K=2 olduğunda verilerin %100, K=3 ve 4 iken %83 ve son olarak K=5 iken %50’sinde en iyi sonuçları yakaladığı görülmektedir. Bu sonuçlar ışığında, CHIO algoritmasının çözüm kalitesi açısından rekabet edici olduğu tespit edilmiştir. Sonuç olarak CHIO algoritması çok düzeyli görüntü eşiği problemi için alternatif bir algoritma olabilir.
Citation - WoS: 1
Covid-19 Detection Using Variational Mode Decomposition of Cough Sounds
(2023) Solak, Fatma Zehra
According to the World Health Organization, cough is one of the most prominent symptoms of the COVID-19 disease declared as a global pandemic. The symptom is seen in 68% to 83% of people with COVID-19 who come to the clinic for medical examination. Therefore, during the pandemic, cough plays an important role in diagnosing of COVID-19 and distinguishing patients from healthy individuals. This study aims to distinguish the cough sounds of COVID-19 positive people from those of COVID-19 negative, thus providing automatic detection and support for the diagnosis of COVID-19. For this aim, “Virufy” dataset containing cough sounds labeled as COVID-19 and Non COVID-19 was included. After using the ADASYN technique to balance the data, independent modes were obtained for each sound by utilizing the Variational Mode Decomposition (VMD) method and various features were extracted from every mode. Afterward, the most effective features were selected by ReliefF algorithm. Following, ensemble machine learning methods, namely Random Forest, Gradient Boosting Machine and Adaboost were prepared to identify cough sounds as COVID-19 and Non COVID-19 through classification. As a result, the best performance was obtained with the Gradient Boosting Machine as 94.19% accuracy, 87.67% sensitivity, 100% specificity, 100% precision, 93.43% F-score, 0.88 kappa and 93.87% area under the ROC curve.
A Deep Learning Ensemble Approach for X-Ray Image Classification
(Konya Teknik Univ, 2024) Esme, Engin; Kıran, Mustafa Servet
The application of deep learning-based intelligent systems for X-ray imaging in various settings, including transportation, customs inspections, and public security, to identify hidden or prohibited objects are discussed in this study. In busy environments, x-ray inspections face challenges due to time limitations and a lack of qualified personnel. Deep learning algorithms can automate the imaging process, enhancing object detection and improving safety. This study uses a dataset of 5094 x-ray images of laptops with hidden foreign circuits and normal ones, training 11 deep learning algorithms with the 10-fold cross-validation method. The predictions of deep learning models selected based on the 70% threshold value have been combined using a meta-learner. ShuffleNet has the highest individual performance with 83.56%, followed by InceptionV3 at 81.30%, Darknet19 at 78.92%, DenseNet201 at 77.70% and Xception at 71.26%. Combining these models into an ensemble achieved a remarkable classification success rate of 85.97%, exceeding the performance of any individual model. The ensemble learning approach provides a more stable prediction output, reducing standard deviation among folds as well. This research highlights the potential for safer and more effective X-ray inspections through advanced machine learning techniques.
Detection of Covid-19 Severity and Mortality From Blood Parameters by Ensemble Learning Methods
(2023) Erol, Doğan, Gizemnur; Uzbaş, Betül
COVID-19 is a pandemic that causes a high rate of spread and Acute Respiratory Distress Syndrome (ARDS). Severe pneumonia in infected individuals has resulted in too many patients being admitted to the Intensive Care Unit (ICU). This has placed unprecedented pressure on health systems by exceeding capacities. It is essential to detect the prognosis of this disease so that the health systems can remain active and the conditions of the patients who need to be hospitalized in the ICU do not become critical. In this study, COVID-19 prognosis was detected by using ICU admission (COVID-19 SEVERITY) and COVID-19 related death (COVID19 MORTALITY) datasets with Machine Learning (ML) methods. The missing data of the datasets were filled with K-Nearest Neighbor (KNN), and Min-Max normalization was performed. Datasets were divided three times into training and test sets, and the data were balanced with the Synthetic Minority Oversampling Technique (SMOTE). Then, classification was carried out using Ensemble Learning (EL) methods. For COVID-19 SEVERITY and COVID-19 MORTALITY, 89.54% and 97.25% accuracy were achieved with the Adaboost classifier, respectively. Successful and rapid COVID-19 prognosis detection with ML methods will help to use the ICU more efficiently and relieve the pressure on health systems.
Determining the Most Powerful Features in the Design of an Automatic Sleep Staging System
(2023) Özşen, Seral; Koca, Yasin; Tezel, Gülay; Çeper, Sena; Küççüktürk, Serkan; Vatansev, Hülya
Spending too much time on manual sleep staging is tiring and challenging for sleep specialists. In addition, experience in sleep staging also creates different decisions for sleep experts. The search for finding an effective automatic sleep staging system has been accelerated in the last few years. There are many studies dealing with this problem but very few of them were conducted with real sleep data. Studies have been carried out on mostly processed and cleaned-ready data sets. In addition, there are few studies in which the data distribution in sleep stages is balanced (equal numbers of epochs from each stage are used), and it is seen that the performance of these studies is quite low compared to other studies. When the literature studies are examined, there is a wide range of studies in which many features are extracted, many feature selection methods are used, many classifiers are applied and various combinations of these are available. For this reason, to determine the best-performing features and the most powerful features, 168 features were extracted from the real EEG, EOG, and EMG signals of 124 patients. These features were selected with 7 different feature selection methods, and classification was carried out with 4 classifiers. In general, the ReliefF feature selection method has performed best, and the Bagged Tree classifier has reached the highest classification accuracy of 67.92% with the use of nonlinear features.

Browse

Browsing Bilgisayar ve Bilişim Fakültesi Koleksiyonu by Department "KTÜN"