Search

Son yıllarda kalp hastalıkların hızla artışı ile birlikte bu hastalıklardan ölen insanların sayısı da hızla artmaktadır. Bu nedenle hastalıkların erken teşhisi, insan yaşam ömrünü arttırmada oldukça önemlidir. Çalışmada 2015 yılı nisan ve aralık aylarını kapsayan kalp yetmezliği hastalarına ait toplam 299 klinik hastaya ait kayıtlar açık erişimli internet sitesinden alınarak kullanılmıştır. Açık erişimli internet sitesinden elde edilen veriler veri ön işleme aşamasından geçirilerek yapay zekâ ile analiz için hazır hale getirilmiştir. Veri seti ilk olarak Rasgele Orman algoritmasının hiper parametreleri belirlenmiştir. Elde edilen hiper parametreler, ikili parçacık sürü zekâ yöntemi kullanılarak veri seti üzerinde global bir özellik seçimi gerçekleştirilmiştir. Elde edilen sonuçlar tekrar rasgele orman algoritması ve 10 kat 10 iterasyonu ile çapraz doğrulama ile eğitilerek ölüm risk durumunu gösteren en uygun model elde edilmiştir. Elde edilen modelin 100 eğitimin ortalama doğruluk oranı ise %79.66 olarak belirlenmiştir., With the rapid increase of heart diseases in recent years, the number of people who die from these diseases is also increasing rapidly. Therefore, early diagnosis of diseases is very important in increasing human life span. In the study, records of 299 clinical patients with heart failure, covering the months of April and December 2015, were taken from the open access website and used. The data obtained from the open access website were made ready for analysis with artificial intelligence by passing through the data preprocessing stage. Firstly, the data set was determined by hyper parameters of the Random Forest algorithm. Using the obtained hyper parameters, binary particle swarm intelligence method, a global feature selection was made from the data set. The obtained results were trained again with random forest algorithm and 10 times 10 iteration cross validation, and the most appropriate model showing death risk situation was obtained. The average accuracy rate of 100 trainings of the model obtained was determined as 79.66%.

Fraud detection identifies suspicious activities, false pretenses, wrongful or criminal deception intended to result in financial gain. Fraud is rare, well thought, effortful, and deceiving throughout claims. Detecting fraudulent claims is essential for the insurance industry. Therefore, most insurance companies must devote time and budget to fraud detection. Fraud detection can be divided into two categories; the main and most common type of fraud is individual fraud. Individual frauds can appear in many kinds of forms. For example, damage to an asset might be occurred before issuing a policy and be reported after. The second category is organized fraud which is much rarer and harder to detect than individual fraud. Especially motor insurance fraud is commonly attempted by organized crime rings. Counterparties involved in fraudulent claims change frequently, and changes make fraud detection difficult. According to Insurance Information and Monitoring Center findings, the fraudulent claim payment ratio is 10 to 30 %, and the detection success rate for an individual is at 1.4 to 5%. At the same time, the annual fraud cost is at 200 to 300 $ million. This study proposes a fraud detection platform called SOBE, which assists fraud departments’ claim inquiry more easily and shorter than manual investigation made by employees. At its core, SOBE uses a rule engine approach. In order to support the rule engine, there is also a machine learning algorithm for fraud detection. In addition, the SNA module detects interconnected fraud counterparts among claim files. Consequently, the SOBE fraud detection platform allows Anadolu Sigorta to prevent improper payments from claiming participants. SOBE platform, the central fraud detection platform at Anadolu Sigorta, was developed in-house using different technologies and methods, including KNIME Analytics Platform, Python, graph methods, and web service methodologies.

Personalizing the e-commerce experience is vital since there are enormous amounts of products to offer customers. Each day new products are introduced into the ecosystem, and customer purchase behavior is dynamic as well. This mapping between products and customers needs to be optimized. E-commerce platforms try to funnel those products by a variety of methods like user clustering and product propensity analysis. The brand propensity metric is one of those key features for personalizing products offered to the customer. Once the brand propensity is calculated, it can be used to cluster customers or list products within the same brand. Since customers periodically interact with different products, these interactions (e.g., product visit, favorite, basket, search, and order) are aggregated to predict the next actions of the corresponding customer. Typically, the next action might be an order action or click. In this study, we develop Logistic Regression (LR) models to investigate the effect of the target variable on calculating brand propensity. For comparison purposes, models based on Decision Tree (DT), Random Forest (RF), and XGBoost (XGB) have also been developed. The target variable to be evaluated for the brand propensity model has been set to both order probability and click probability. The “Top N accuracy” metric has been used to evaluate the performance of the models. As the study’s outcome, click as a target variable has been revealed to be more beneficial since it also shows that customers are more likely to explore what is inside that brand. In addition, the LR-based propensity models exhibit the best average performance for both Top 3 and Top 5 accuracies among the machine learning methods.

This study approaches the least-squares method for simple linear regression model. The least-squares line does not comply with the data when there are outliers that have deceptive effects on the results in the dataset. The study aims to develop a method for obtaining a line that complies more with the data when there are outliers in the dataset.

Nowadays, trade based on logistics and sea transportation has gained importance. Considering the traffic related to this, the classification and discrimination of ship types are important in terms of transportation and storage costs, and safety. The classification of ships performing different tasks on the sea is handled in this study, and a ship image dataset has been created in order to make a high accuracy ship classification thanks to deep learning methods. It is preferred in our deep learning study in comparison to the classical machine learning method, as the features are semantically richer as the higher level, while expressing content from the dataset. It was created by the acquisition of various ship images thanks to the web scraping method. YOLOv5 and Xception deep learning models were trained to obtain the most appropriate classification performance. As a result of the experiments, an accuracy rate of approximately between 96% and 99% was achieved with both models. Scientific findings and discussion are included in our study as well., Günümüzde lojistik ve deniz ulaşımına dayanan ticaret oldukça önem kazanmıştır. Buna dair oluşan trafik göz önüne alındığında gemi çeşitlerinin sınıflandırılarak ayrıştırılması, taşıma, depolama maliyetleri açısından ve güvenlik konusunda önem arz etmektedir. Deniz üzerinde farklı görevleri icra etmekte olan gemilerin sınıflandırılması bu çalışmada ele alınarak, derin öğrenme yöntemleri sayesinde yüksek doğrulukta bir gemi sınıflandırma yapılabilmesi için gemi görüntüleri veri kümesi oluşturulmuştur. Veri kümesinden elde edilen içeriği ifade ederken özniteliklerin daha yüksek seviyeden anlamsal olarak zengin olmasından dolayı, klasik makine öğrenmesi yöntemine kıyasla derin öğrenme çalışmamızda tercih edilmiştir. Derin öğrenme modellerinin eğitiminde ve test edilmesinde kullanılmak üzere bu veri kümesi açık kaynaklı Internet sitelerinden ağ kazıma (web scraping) yöntemi sayesinde çeşitli gemi görüntülerinin edinimi ile oluşturulmuştur. YOLOv5 ve Xception derin öğrenme modelleri eğitilerek en uygun sınıflandırma başarımı elde edilmiştir. Deneyler sonucunda her iki model ile yaklaşık olarak %96 ilâ %99 arası doğruluk oranında başarım değerine ulaşılmıştır. Varılan bilimsel bulgulara ve tartışmaya çalışmamızda yer verilmektedir.

Semantic segmentation is a fundamental problem for computer vision. On the other hand, for studies in the field of biometrics, semantic segmentation is gaining more importance. Many successful biometric recognition systems require a high- performance semantic segmentation algorithm. In this study, we present an effective ear segmentation technique in natural images. A convolutional neural network is trained for pixel-based ear segmentation. DeepLab v3+ network structure, with ResNet-18 as the backbone and Tversky lost function layer as the last layer, has been trained with natural and uncontrolled images. We perform the proposed network training using only the 750 images in the Annotated Web Ears (AWE) training set. The corresponding tests are performed on the AWE Test Set, University of Ljubljana Test Set, and the Collection A of In-The-Wild dataset. For the Annotated Web Ears (AWE) dataset, intersection over union (IoU) is measured as 86.3% for the AWE database. To the best of our knowledge, this is the highest performance achieved among the algorithms tested on the AWE test set.

Drowsiness is one of the major causes of driver-induced traffic accidents. The interactive systems developed to reduce road accidents by alerting drivers is called as Advanced Driver Assistance Systems (ADAS). The most important ADAS are Lane Departure Warning System, Front Collision Warning System and Driver Drowsiness Systems. In this study, an ADAS system based on eye state detection is presented to detect driver drowsiness. First, Viola-Jones algorithm approach is used to detect the face and eye areas in the proposed method. The detected eye region is classified as closed or open by making use of a machine learning method. Finally, the eye conditions are analyzed at time domain with PERcentage of eyelid CLOsure (PERCLOS) metric and drowsiness conditions are determined by Support Vector Machine (SVM), kNN and decision tree classifiers. The proposed methods tested on 7 real people and drowsiness states are detected at 99.77%, 94.35%, and 96.62% accuracy, respectively.

Finding the minimum distance of linear codes is a non-deterministic polynomial-time-hard problem and different approaches are used in the literature to solve this problem. Although, some of the methods focus on finding the true distances by using exact algorithms, some of them focus on optimization algorithms to find the lower or upper bounds of the distance. In this study, we focus on the latter approach. We first give the swarm intelligence background of artificial bee colony algorithm, we explain the algebraic approach of such algorithm and call it the algebraic artificial bee colony algorithm (A-ABC). Moreover, we develop the A-ABC algorithm by integrating it with the algebraic differential mutation operator. We call the developed algorithm the mutation-based algebraic artificial bee colony algorithm (MBA-ABC). We apply both; the A-ABC and MBA-ABC algorithms to the problem of finding the minimum distance of linear codes. The achieved results indicate that the MBA-ABC algorithm has a superior performance when compared with the A-ABC algorithm when finding the minimum distance of Bose, Chaudhuri, and Hocquenghem (BCH) codes (a special type of linear codes).

Partikül madde (PM) kirliliği önemli çevresel sorunlara sebep olmaktadır. PM kirliliğinin olumsuz etkileri, canlı sağlığına yönelik riskleri nedeniyle yaygın bir sorun haline gelmiştir. PM kirliliğinin tüm bu olumsuz etkileri ve atmosferdeki karmaşık etkileşimi sebebiyle, daha fazla çalışmaya konu olması önemlidir. Özellikle, PM kirliliğinin izlenmesi ve tahmin edilmesi konusunda yapılacak çalışmalar önemlidir. Son yıllarda meteorolojik faktörler göz önüne alınarak PM kirliliğinin tahmin edilmesi çalışmaları artmıştır. Özellikle makine öğrenme yöntemleri ile PM kirliliği tahmini çalışmaları hız kazanmıştır. Bu çalışmada, meteorolojik faktörler göz önüne alınarak çeşitli makine öğrenme algoritmaları ile PM10 kirliliği tahmin edilmiştir. Çalışmada kullanılan meteoroloji verileri Meteoroloji Genel Müdürlüğü Ankara Bölge istasyonundan (enlem:39,9727, boylam:32,8637, rakım:891 m.) elde edilmiştir. PM10 kirlilik verileri ise Çevre, Şehircilik ve İklim Değişikliği Bakanlığı Ankara Keçiören-Sanatoryum hava kalitesi istasyonundan (enlem: 39,999, boylam: 32,856, rakım: 1009 m.) elde edilmiştir. Makine öğrenme çalışması aşamasında, sıcaklık, çiğ noktası sıcaklığı, yağış, bağıl nem, rüzgar hızı, basınç, bulut kapalılığı ve bir önceki güne ait PM10 ölçümleri göz önüne alınarak, farklı makine öğrenme (karar ağacı regresyonu, destek vektör regresyonu, lasso regresyonu ve yapay sinir ağı) algoritmalarıyla ayrı ayrı çalışma yapılmış ve bu algoritmaların tutarlılıkları karşılaştırılmıştır. Tutarlılıklarının incelenmesi aşamasında çeşitli istatistiksel metrikler kullanılmıştır. Sonuçta, test bölümü göz önüne alındığında, yapay sinir ağı algoritmasının belirleme katsayısı ̴0,6, kök ortalama kare hatası ̴18 ve ortalama mutlak hata ̴12 olarak bulunmuş ve yapay sinir ağı algoritmasının diğer algoritmalara göre daha iyi sonuç verdiği görülmüştür., Particulate matter (PM) pollution causes significant environmental problems. The adverse effects of PM pollution have become a common problem due to its risks to living health. Due to all these negative effects of PM pollution and its complex interaction in the atmosphere, it is important that it be the subject of more studies. In particular, studies on monitoring and estimating PM pollution are important. In recent years, studies on estimating PM pollu-tion have increased by considering meteorological factors. Especially with machine learning methods, PM pollution estimating has accelerated. In this study, PM10 pollution is estimated with various machine learning algorithms considering meteorological factors. The meteorological data used in the study were obtained from the Ankara Regional Station of Turkish State Meteorological Service (latitude: 39,9727, longitude: 32,8637, altitude: 891 m.). PM10 pollution data were obtained from the Ministry of Environment, Urbanization and Climate Change Ankara Keçiören-Sanatorium air quality station (latitude: 39,999, longitude: 32,856, altitude: 1009 m.). In the machine learning phase, different machine learning (decision tree regression, support vector regression, lasso regression and neural network) were used, considering temperature, dew point temperature, precipitation, relative humidity, wind speed, pressure, cloud cover and PM10 measurements of the previous day. Algorithms were studied separately and the consistencies of these algorithms were compared. Various statistical metrics were used to examine their con-sistency. As a result, considering the test section, the determination coefficient was found to be ̴0,6, root mean square error ̴18, and mean absolute error ̴12 for artificial neural network algorithm, and it was seen that the artifi-cial neural network algorithm gave better results than other algorithms.

The aim of this study is to determine minimum quantity of unmanned aerial vehicles (UAVs) that should be used in an area where an aerial reconnaissance/observation activity will be carried out. In order to make this kind of calculation, firstly, the energy consumption of a UAV while flying with a constant speed was examined and then % energy level consumed by the UAV while passing each meter was obtained approximately. In this study, by considering the length of the trajectory which a UAV will navigate, required % energy level of a UAV to complete a single tour is calculated. If it is determined that one UAV can not complete its assigned trajectory, the number of UAVs are increased until each UAV complete its trajectory. In this study, the vehicle routing problem approach was used to calculate the UAV trajectories. Genetic algorithm method that is one of the metaheuristic optimization methods, was used obtain the solution of the vehicle routing problem (VRP). The developed algorithm has been run in Matlab environment. By changing the parameters of crossing rate and the population number in the genetic algorithm (GA) method, the lowest number of UAVs that is enough to carry out aerial observation over a target area and the shortest UAV trajectories were obtained. Results were presented in table form.

Epilepsy is one the most prevalent neurological disorders whose causes are not exactly known. Diagnosis and treatment of epilepsy are closely related to the patient's story, and the most important indicator is the frequency and severity of seizures. Since the disease does not only affect the patients but also the lives of their environment seriously, it is very important to make the diagnosis and treatment correctly. However, sometimes misrecognition from patients and their relatives, unnecessary epilepsy treatment to the patient in non-epileptic seizures mixed with epileptic seizures, or increasing the dose of the drugs used for the patient are the situations frequently encountered. The so-called video-EEG method is used in the detection and segregation of epileptic / non-epileptic seizures. In this method, the patient is kept in an environment where video recording is continuously taken until the seizure occurs, and EEG, EMG, and ECG records of the patient are taken. When the patient has a seizure, the seizure type is separated by examining these records. In this project, seizure detection and seizure type (epileptic / non-epileptic) detection is aimed to be done by using wearable sensors increasingly applied in the field of health. The achievable benefits from the project and data set will provide a different perspective on the epilepsy illness, as well as reduce the number of epilepsy patients who are not in fact epilepsy patients needing treatment, and keep epileptic seizure recordings constantly in the electronic environment so that the treatment processes are monitored more closely.

Confidentiality requirements of individuals and companies led to the dominance of encrypted payloads in the overall Internet traffic. Hence, traffic classification on a network became increasingly difficult as it must rely on only the packet headers. Many vital tasks such as differential pricing, providing a safe Internet for children, and eliminating malicious connections require traffic classification, even if the payload contents are encrypted. Encrypted traffic is harder to classify as packet content becomes unreadable. In this work, we aim to provide an insight into traffic classification using encrypted packets in terms of both accuracy and packet processing time. LSTM (Long Short-Term Memory) architecture is a good candidate for this problem as it can handle sequences. Each flow can be modeled as a sequence and patterns of the sequences can provide valuable information. We compare the performance of LSTM with other methods in both real-time and offline experiments. Compared to a machine learning method both online and offline LSTM excelled with precision and recall differences up to 50%. Average accuracy with LSTM was measured as 97.77% offline and 91.7% in real-time. Average packet processing time in real-time was recorded as 0.593 msec which is 5 times faster than a recent work that uses LSTM method.

Electromyography (EMG) signals have been extensively used for identification of finger movements, hand gestures and physical activities. In the classification of EMG signals, the performance of the classifier is widely determined by the feature extraction methods. Thus, plenty of feature extraction methods based on time, histogram and frequency domain have been reported in literature. However, these methods have several drawbacks such as high time complexity, high computation demand and user supplied parameters. To overcome these deficiencies, in this work, a new feature extraction method has been proposed to classify EMG signals taken from two different data sets finger movements (FM) and physical actions (PA). While FM data set includes 14 different finger movements, PA data set involves 20 different physical activities. The proposed method is based on numerical fractional integration of time series EMG signals with different fractional-orders. K Nearest Neighborhood (KNN) classifier with 8-fold cross validation has been employed for prediction of EMG signals. The derived fractional features can give better results than the two commonly used time domain features, notably, mean absolute value (MAV) and waveform length (WL) in terms of accuracy. The experimental results are also supported by statistical analysis results.

A new hybrid machine learning method for the prediction of type 2 diabetes is introduced and explained in detail. Also, outcomes are compared with similar researches. Early prediction of diabetes is crucial to take necessary measures (i.e. changing eating habits, patient weight control etc.), to defer the emergence of diabetes and to reduce the death rate to some extent and ease medical care professionals’ decision-making in preventing and managing diabetes mellitus. The purpose of this study is the creation of a new hybrid feature selection approach combination of Correlation Matrix with Heatmap and Sequential forward selection (SFS) to reveal the most effective features in the detection of diabetes. A diabetes data set with 520 instances and seven features were studied with the application of the proposed hybrid feature selection approach. The evaluation of the selected optimal features was measured by applying Support Vector Machines(SVM), Random Forest(RF), and Artificial Neural Networks(ANN) classifiers. Five evaluation metrics, namely, Accuracy, F-measure, Precision, Recall, and AUC showed the best performance with ANN (99.1%), F-measure (99.1%), Precision (99.3%), Recall (99.1%), and AUC (99.2%). Our proposed hybrid feature selection model provided a more promising performance with ANN compared to other machine learning algorithms.

Pneumonia is a general public health problem. It is an important risk factor, especially for children under 5 years old and people aged 65 and older. Fortunately, it is a treatable disease when diagnosed in the early phase. The most common diagnostic method known for the disease is chest X-Rays. However, the disease can be confused with different disorders in the lungs or its variants by experts. In this context, computer-aided diagnostic systems are necessary to provide a second opinion to experts. Convolutional neural networks are a subfield in deep learning and they have demonstrated success in solving many medical problems. In this paper, Xception which is a convolutional neural network was trained with the transfer learning method to detect viral pneumonia, normal cases, and bacterial pneumonia in chest X-Rays. Then, five different machine learning classification algorithms were trained with the features obtained by the trained convolutional neural network. The classification performances of the algorithms were compared. According to the test results, Xception achieved the best classification result with an accuracy of 89.74%. On the other hand, SVM achieved the closest classification performance to the convolutional neural network model with 89.58% accuracy.

Identifying subgroups of cancer patients is important as it opens up possibilities for targeted therapeutics. A widely applied approach is to group patients with unsupervised clustering techniques based on molecular data of tumor samples. The patient clusters are found to be of interest if they can be associated with a clinical outcome variable such as the survival of patients. However, these clinical variables of interest do not participate in the clustering decisions. We propose an approach, WSURFC (Weakly Supervised Random Forest Clustering), where the clustering process is weakly supervised with a clinical variable of interest. The supervision step is handled by learning a similarity metric with features that are selected to predict this clinical variable. More specifically, WSURFC involves a random forest classifier-training step to predict the clinical variable, in this case, the survival class. Subsequently, the internal nodes are used to derive a random forest similarity metric among the pairs of samples. In this way, the clustering step utilizes the nonlinear subspace of the original features learned in the classification step. We first demonstrate WSURFC on hand-written digit datasets, where WSURFC is able to capture salient structural similarities of digit pairs. Next, we apply WSURFC to find breast cancer subtypes using mRNA, protein, and microRNA expressions as features. Our results on breast cancer show that WSURFC could identify interesting patient subgroups more effectively than the widely adopted methods.

Altın, en çok taklit edilen değerli metallerden biridir. Bakırın rengi altına benzer. Bu nedenle bakır, renk sahteciliği için en yaygın kullanılan malzemelerden biridir. Kimyasal özellikler söz konusu olduğunda, volfram altına benzer (altın ve tungstenin yoğunluğu sırasıyla 19.30 g/ml ve 19.25 g/ml'dir), bu nedenle kimyasal bir sahte olarak kullanılabilir. Altının saflığı X-ray ile belirlenebilir, ancak bu yöntem maliyetlidir. Bu yazıda, sahte altın tespiti için kuyumcuların mevcut düşük maliyetli yöntemleri ve sahte parayı tespit etmek için kullanılan düşük maliyetli yöntemler denenmiştir. Bir yüzeye altın bir madde çarptığında, ses frekansı aynı deney bakır ile yapıldığındaki sesin frekansından daha yüksektir. Ayrıca, sahte altın rengi gerçek olanlardan daha parlaktır. Altın rengi benzersizdir ve "altın sarısı" olarak adlandırılır. Bu araştırmada ses ve görüntü işleme yöntemleri kullanılarak sahte ve orijinal altın ayrımı yapılmıştır. Görüntü işleme kısmı için, önce görüntüden altını segmentlere ayırmak için CNN tabanlı bir araç kutusu uygulanır. Bundan sonra, altın ve bakır malzemelerin rengini ayırt etmek için derin Evrişimli Sinir Ağları kullanılır. Hem ses hem de görüntü işleme teknikleri ile umut verici sonuçlar elde edilmektedir., Gold is one of the most counterfeited precious metals. The color of copper is like gold. For this reason, copper is one of the most used materials for color counterfeiting. When the chemical properties are concerned, wolfram is like gold (density of gold and tungsten are 19.30 g/ml and 19.25 g/ml, respectively), so it can be used as a chemical counterfeit. The purity of gold can be determined by X-ray, but this method is costly. The current low-cost methods of jewelers have been experimented with for counterfeit gold detection in this paper. When a gold matter is hit by a subject, the sound frequency is higher than the frequency of sound when the same experiment is done with copper. Furthermore, counterfeit gold color is brighter than real ones. The color of gold is unique, and it is called "gold yellow". In this research, by employing sound and image processing, counterfeit and original gold are differentiated. For the image processing part, first a Convolutional Neural Network (CNN)-based toolbox for segmenting the gold material is applied. Then, deep CNNs for differentiating the color of the gold and copper materials are employed. Promising results are achieved with both sound and image processing techniques.

Kanser dünya genelinde pek çok insanın ölümüne sebep olan en önemli hastalıklardan biridir. Özellikle göğüs kanseri kadınlar arasında en çok rastlanan hastalıkların başında yer almaktadır. Bu sebeple kanser hastalığının teşhisi ile alakalı herhangi bir gelişme insanların sağlıklı bir yaşam sürmesi açısından oldukça önemlidir. Günümüzde makine öğrenmesi yöntemlerinin kullanılması, kanser hastalığının erken teşhisi ve tahmini için yapılan çalışmalara büyük katkılar sağlamaktadır. Bu çalışmada da k-En Yakın Komşu, Destek Vektör Makinaları, Naive Bayes, Karar ağaçları ve Yapay Sinir Ağları gibi beş farklı makine öğrenmesi yöntemleri Kaggle platformunda yer alan iki farklı göğüs kanseri veri kümesi üzerinde uygulanmıştır. Elde edilen sonuçlar doğruluk değerleri ve karmaşıklık matrisi değerleri ile verilerek karşılaştırılmıştır. Birinci göğüs kanseri veri kümesi içinde %98,2456 doğruluk oranıyla ve ikinci göğüs kanseri veri kümesinde %93,8596 doğruluk oranıyla Yapay Sinir Ağları (YSA) yönteminde en yüksek doğruluk değerleri elde edilmiştir., Cancer is one of the most important diseases that cause the death of many people around the world. Especially, breast cancer is one of the most common diseases among women. For this reason, any development related to the diagnosis of cancer is critical for people to live healthy lives. Today, the use of machine learning methods makes great contributions to studies for the early diagnosis and prediction of cancer disease. In this study, five different machine learning methods such as k-Nearest Neighbor, Support Vector Machines, Naive Bayes, Decision Trees, and Artificial Neural Networks were applied on two other breast cancer datasets on the Kaggle platform. The obtained results were compared by giving accuracy values and confusion matrix values. The highest accuracy values were obtained in Artificial Neural Networks (ANN) method with an accuracy rate of 98.2456% in the first breast cancer dataset and 93.8596% in the second breast cancer dataset.

Education is the foundation of economic, social, and cultural development for every individual and society as a whole. Students are accepted to secondary education institutions with the high school entrance examination made by the Ministry of National Education in Turkey. In this study, the success rates of the students who took the high school entrance examination in Turkey's 81 provinces in 2019 were handled with the machine learning regression and beta regression model. The present paper aimed to model, predict, and explain students' success rates using variables such as divorce rate, gross domestic product, illiteracy, and higher education populations. Support vector regression, random forest, decision tree, and beta regression model were applied to estimate success rates. Two models with the highest R2 value were found to be beta regression and random forest models. When the prediction errors of beta regression and random forest model were examined, it seemed to be that the random forest model is relatively superior to the beta regression model in predicting the success rates. While the beta regression model was the best predictor of the success rates of Çanakkale province, the random forest model predicted the success rates of Ankara well. Also, it was seen that the variables found to be significant in the beta regression model for success rates were also crucial in the random forest model. It is recommended to use both the beta and random forest models to estimate the students' success rates., Eğitim, her birey ve bir bütün olarak toplum için ekonomik, sosyal ve kültürel gelişimin temelidir. Ülkemizde orta öğretim kurumlarına Milli Eğitim Bakanlığı tarafından yapılan lise giriş sınavı ile öğrenci kabul edilmektedir. Bu çalışmada, 2019 yılında Türkiye'nin 81 ilinde lise giriş sınavına giren öğrencilerin başarı oranları makine öğrenimi regresyon ve beta regresyon modeli ile ele alınmıştır. Bu makale, boşanma oranı, gayri safi yurtiçi hasıla, okuma yazma bilmeyenlerin sayısı ve yüksek öğretim nüfusu sayısı gibi değişkenleri kullanarak öğrencilerin başarı oranlarını modellemeyi, tahmin etmeyi ve açıklamayı amaçlamaktadır. Başarı oranlarını tahmin etmek için destek vektörü regresyon, rastgele orman, karar ağacı ve beta regresyon modeli uygulanmıştır. En yüksek R2 değerine sahip iki modelin beta regresyon ve rastgele orman modelleri olduğu bulunmuştur. Beta regresyon ve rastgele orman modelinin tahmin hataları incelendiğinde, başarı oranlarını tahmin etmede rastgele orman modelinin beta regresyon modeline göre nispeten üstün olduğu görülmektedir. Beta regresyon modeli Çanakkale ilinin başarı oranlarının en iyi yordayıcısı iken, rastgele orman modeli Ankara'nın başarı oranlarını iyi tahmin etmiştir. Ayrıca beta regresyon modelinde başarı oranları için anlamlı bulunan değişkenlerin rastgele orman modelinde de önemli olduğu görülmüştür. Öğrencilerin başarı oranlarını tahmin etmek için hem beta hem de rastgele orman modellerinin kullanılması önerilir.

Customer feedback is one of the most critical parameters that determine the market dynamics of product development. In this direction, analyzing product-related complaints helps sellers to identify the quality characteristics and consumer focus. There have been many studies conducted on the design of Machine Learning (ML) systems to address the causes of customer dissatisfaction. However, most of the research has been particularly performed on English. This paper contributes to developing an accurate categorization of customer complaints about package food products, written in Turkish. Accordingly, various ML algorithms using TF-IDF and word2vec feature representation strategies were performed to determine the category of complaints. Corresponding results of Linear Regression (LR), Naive Bayes (NB), k Nearest Neighbour (kNN), Support Vector Machine (SVM), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) classifiers were provided in related sections. Experimental results show that the best-performing method is XGBoost with TF-IDF weighting scheme and it achieves %86 F-measure score. The other considerable point is word2vec based ML classifiers show poor performance in terms of F-measure compared to the TF-IDF term weighting scheme. It is also observed that each experimented TF-IDF based ML algorithm gives a more successful prediction performance on the optimal subsets of features selected by the Chi Square (CH2) method. Performing CH2 on TF-IDF features increases the F-measure score from 86% to 88% in XGBoost.

The IoT is a sensors world that detects countless physical events in our environment and transforms them into data, and transfers this data to different environments or digital systems. The usage areas of Internet of things-based technologies are constantly increasing and technologies are being developed to support the IoT infrastructure. But, in order to effectively manage the large number of big-data generate in the detection layer, it should be pre-processed and done in accordance with big-data standards. For the effective management of big data, it is imperative to improving the standards of the data set, and filtering methods are being developed for a higher quality data set. For instance, using data cleaning methods is a preprocessing method that facilitates data mining operations. In this way, more manageable data is obtained by preventing the formation of interference and big data can be managed more effectively. In this study, we investigate the efficient operation of IoT and big data originating from the internet of things. Additionally, real-time anomalous data filtering is performed on IoT edges with a data set consisting of six different data produced in real- time. Furthermore, the speed and accuracy performances of classifiers are compared, and machine learning algorithms such as the random cut forest-RCF, logistic regression-LR, naive bayes-NB, and neural network-NN classifiers are used for comparison. According to the accuracy performance values, the RCF and LR classifiers are very close, but considering the speed values, it is seen that the LR classifier is more successful in IoT systems.

Path planning is an essential topic of robotics studies. Robotic researchers have suggested some methods such as particle swarm optimization, A*, and reinforcement learning (RL) to obtain a path. In the current study, it was aimed to generate RL-based safe path planning for a 3R planar robot. For this purpose, firstly, the environment was performed. Later, state, action, reward, and terminate functions were determined. Lastly, actor and critic artificial neural networks (ANN), which are basic components of deep deterministic policy gradients (DDPG), were formed in order to generate a safe path. Another aim of the current study was to obtain an optimum actor ANN. Different ANN structures that have 2, 4, and 8-layers and 512, 1024, 2048, and 4096-units were formed to get an optimum actor ANN. These formed ANN structures were trained during 5000 episodes and 200 steps and the best results were obtained by 4-layer, 1024, and 2048-units structures. Owing to this reason, 4 different ANN structures were performed utilizing 4-layer, 1024, and 2048-units. The proposed structures were trained. The NET-M2U-4L structure generated the best result among 4 different proposed structures. The NET-M2U-4L structure was tested by using 1000 different scenarios. As a result of the tests, the rate of generating a safe path was calculated as 93.80% and the rate of colliding to the obstacle was computed as 1.70%. As a consequence, a safe path was planned and an optimum actor ANN was obtained for a 3R planar robot.

With the introduction of artificial intelligence into our lives, artificial intelligence researches and applications in different fields such as agriculture, health, military and engineering applications have become very popular iris flower was classified using the popular Random Forest, support vector machine and Artificial neural network machine learning classifiers with high accuracy rates. As a result of the classification, the performance of the trained models was evaluated according to the confusion matrix, sensitivity, specificity, accuracy, F1 score, ROC curve and AUC evaluation criteria. The random forest algorithm was the most successful among the trained algorithms with an accuracy rate of 97%.

Convolutional neural networks (CNNs) are one of the popular deep learning methods used to solve the hyperspectral image classification (HSIC) problem. CNN has a strong feature learning ability that can ensure more distinctive features for higher quality HSIC. The traditional CNN-based methods mainly use the 2D CNN for HSIC. However, with 2D CNN, only spatial features are extracted in HSI. Good feature maps cannot be extracted from spectral dimensions with the use of 2D CNN alone. By using 3D CNN, spatial-spectral features are extracted simultaneously. However, 3D CNN is computationally complex. In this study, a hybrid CNN method, which is a combination of 3D CNN and 2D CNN, is improved to solve the two problems described above. Using hybrid CNN decreases the complexity of the method compared to using only 3D CNN and can perform well against a limited number of training samples. On the other hand, in Hybrid CNN, depthwise separable convolution (DSC) is used, which decreases computational cost, prevents overfitting and enables more spatial feature extraction. By adding DSC to the developed hybrid CNN, a hybrid depthwise separable convolutional neural network is obtained. Extensive applications on frequently used HSI benchmark datasets show that the classification performance of the proposed network is better than compared methods.

Nowadays, the scope of machine learning and deep learning studies is increasing day by day. Handwriting recognition is one of the examples in daily life for this field of work. Data storage in digital media is a method that almost everyone is using nowadays. At the same time, it has become a necessity for people to store their notes in digital media and even take notes directly in the digital environment. As a solution to this need, applications have been developed that can recognize numbers, characters, and even text from handwriting using machine learning and deep learning algorithms. Moreover, these applications can recognize numbers, characters, and text from handwriting and convert them into visual characters. This project, investigated the performance comparison of machine learning algorithms commonly used in handwriting recognition applications and which of them are more efficient. As a result of the study, the accuracy was 98.66% with artificial neural network, 99.45% with convolutional neural network, 97.05% with K-NN, 83.57% with Naive Bayes, 97.71% with support vector machine and 88.34% with decision tree. This study also developed a handwriting recognition system for numbers similar to these mentioned applications. A desktop application interface was developed for end users to show the instant performance of some of these algorithms and allow them to experience the handwriting recognition system.

The purpose of occupational health and safety studies is to protect employees from work accidents and occupational diseases and to ensure that they work in a healthy environment. Most of the work accidents happen as a result of wrong storage, transportation and use of chemicals. In order to protect employees from chemical hazards and eliminate their possible risks, a risk assessment should first be carried out. Based on the results of the risk assessment, if not done before, bottles that contain hazardous chemicals must be classified and labelled according to their risk levels. The labels of bottles that contain chemical liquids must be checked and if the labels are worn or unreadable, they must be renewed. After these have been done, hazardous chemical liquids must be classified, stored and transported according to these labels. In this study, a non-contact, liquid measurement system based on microwave data is proposed to detect hazardous liquids. In order to select the most suitable algorithm for use in this measurement system, 3 different classification algorithms have been used and the performance analysis of the algorithms has made. In the study, 3 different classification processes have been applied according to the chemical properties of the liquids. It has been observed that Random Tree algorithm has achieved the best performance while Rep Tree algorithm has done the worst performance. Using this system, hazardous chemical liquids can be detected without opening the cover of the bottles that contain the liquids. Therefore, it can be used to quickly label hazardous liquids for their safe storage and transportation.

Accurate determination of temporal dependencies among gene expression patterns is crucial in the assessment of functions of genes. The gene expression series generally show a periodic behavior with nonlinear curved patterns. This paper presents the determination of temporally associated budding yeast gene expression series by using compositional correlation method. The results show that the method is capable of determining real direct or inverse linear, nonlinear and monotonic relationships between all gene pairs. Pearson’s correlation values between some of the gene pairs have shown negative or very weak relationships (r ≈ 0) even though they were found to be strongly associated. Inversely, a high positive r value was obtained even though the genes are inversely related as determined by the compositional correlation approach. Comparisons with Pearson’s correlation, Spearman’s correlation, distance correlation and the simulated annealing genetic algorithm maximal information coefficient (SGMIC) have shown that the presented compositional correlation method detects important associations which were not found by the compared methods. Supplementary materials containing the code of the used software together with some extended figures and tables are available online.

Classification of the artwork according to their genres is being done for years. Although this process was used to be done by art experts before, now artificial intelligence techniques may help people manage this classification task. The algorithms used for classification are already improved, and now they can make classifications and predictions for any kind of genre classification. In this study, two different machine learning algorithms are used on an artwork dataset for genre classification. The primary purpose of this study is to show that the derived features about the artwork have a remarkable effect on correct genre classification. These features are derived from the metadata of the dataset. This metadata contains information about the nationalities and the period that the artist lived. Image filters are also applied to the images but the results show that applying only image filters on the dataset used in the study did not perform well. Instead, adding derived features extracted from the metadata increased the classification performances dramatically.

Deep learning researchers and practitioners have accumulated a significant amount of experience on training a wide variety of architectures on various datasets. However, given a network architecture and a dataset, obtaining the best model (i.e. the model giving the smallest test set error) while keeping the training time complexity low is still a challenging task. Hyper-parameters of deep neural networks, especially the learning rate and its (decay) schedule, highly affect the network's final performance. The general approach is to search the best learning rate and learning rate decay parameters within a cross-validation framework, a process that usually requires a significant amount of experimentation with extensive time cost. In classical cross-validation (CV), a random part of the dataset is reserved for the evaluation of model performance on unseen data. This technique is usually run multiple times to decide learning rate settings with random validation sets. In this paper, we explore batch-level cross-validation as an alternative to the classical dataset-level, hence macro, CV. The advantage of batch-level or micro CV methods is that the gradient computed during training is re-used to evaluate several different learning rates. We propose an algorithm based on micro CV and stochastic gradient descent with momentum, which produces a learning rate schedule during training by selecting a learning rate per epoch, automatically. In our algorithm, a random half of the current batch (of examples) is used for training and the other half is used for validating several different step sizes or learning rates. We conducted comprehensive experiments on three datasets (CIFAR10, SVHN and Adience) using three different network architectures (a custom CNN, ResNet and VGG) to compare the performances of our micro-CV algorithm and the widely used stochastic gradient descent with momentum in a early-stopping macro-CV setup. The results show that, our micro-CV algorithm achieves comparable test accuracy to macro-CV with a much lower computational cost.

Many network designs in recent years have offered deeper layered solutions. However, models that achieve high-performance results with fewer layers are preferred due to causing less processing load for the system. The U-Net authors succeeded in efficiently creating a model with fewer layers. However, the U-Net architecture also requires improvement to become more efficient. For this purpose, we offer a novel encoder-decoder architecture based on the U-Net and the LU-Net. Furthermore, we propose using a reduced number of up-sampling operations, which were utilized together with the down-sampling operations intensively in the encoder section in our previous research, in the encoder part. The proposed architecture was evaluated on the IOSTAR dataset for the segmentation of retinal vessels. The preprocessing and data augmentation processes were applied to the images before training. The U-Net, LU-Net, and the proposed model were evaluated by using the accuracy, sensitivity, specificity, Dice, and Jaccard metrics. The proposed model achieved performance metric values such as an accuracy of 97.29%, a sensitivity of 81.10%, a specificity of 98.94%, a Dice coefficient of 84.66%, and a Jaccard coefficient of 73.41%. The proposed model obtained improved results compared with the other models, especially for test samples.

The coronavirus pandemic, which emerged at the end of 2019, continues to be effective. Although variousvaccines have been developed, the diagnosis of the disease and the 14-day isolation method still remain valid.Purpose: To classify diseases such as COVID-19, tuberculosis, other pneumonias, and lung opacity. For this goal, itwas aimed to automate disease diagnosis, which was done manually by expert radiologists with chest X-Ray images,with a hybrid model created with the four most commonly used convolutional neural networks. In addition, it is aimedto reach a more accurate result by combining four CNN models instead of one model during the epidemic whendiagnosis is of critical importance. Method: In the binary classification, the images were categorised as COVID-19-positive and negative, and multi-class classification was conducted according to labels including Lung Opacity,COVID-19, Normal, and Viral Pneumonia. The recommended model consisted of MobileNetV2, DenseNet121,InceptionResNetV2, and Xception networks. Transfer learning was used for initial weights. These models, werecombined with ensemble learning to obtain better classification performance. Results: The best binary classificationresult was obtained from the MobileNetv2 with an accuracy rate of 98.84%. The hybrid model improved slightly,with a value 0.16. In the multi-class classification, the DenseNet121 accomplished the highest classification rate, with93.17%. The hybrid model improvement in the multi-class classification was 0.81. According to these results, theproposed method will alleviate the burden of health personnel during the epidemic. It will also aid in the diagnosisof COVID-19 in areas where facilities are limited.

Activation function is a parameter that directly affects the success of the neural network. Generally, nonlinear activation functions are used to solve complex problems. Logistic sigmoid function and tangent hyperbolic function are the most commonly used activation functions. The presence of a free parameter in the activation function makes the function adaptive. In this study, generalized sigmoid-based adaptive activation function was investigated and compared with standard logistic sigmoid and tangent hyperbolic function in fully connected and convolutional neural network architectures. The experimental results show that the network is trained faster thanks to the free parameters, but it is more prone to overfitting., Sinir ağının başarısını aktivasyon fonksiyonu doğrudan etkileyen bir parametredir. Genellikle lineer olmayan aktivasyon fonksiyonlar karmaşık problemleri çözebilmek için kullanılır. Lojistik sigmoid fonksiyon ve tanjant hiperbolik fonksiyon en yaygın kullanılan aktivasyon fonksiyonlarındandır. Aktivasyon fonksiyonunda serbest parametrenin bulunması fonksiyonun adaptif olmasını sağlar. Bu çalışmada genelleştirilmiş sigmoid tabanlı adaptif aktivasyon fonksiyonu incelenmiş ve tam bağlı ve konvolüsyonel sinir ağ mimarili yapılarda standart lojistik sigmoid ve tanjant hiperbolik fonksiyonla kıyaslanmıştır. Elde edilen deneysel sonuçlar, serbest parametreler sayesinde ağın daha hızlı eğitildiğini ancak aşırı uyumlamaya daha yatkın olduğunu göstermektedir.

People mostly communicate through speech or facial expressions. People's feelings and thoughts are reflected in their faces and speech. This phenomenon is an important tool for people to empathize when communicating with each other. Today, human emotions can be recognized automatically with the help of artificial intelligence systems. Automatic recognition of emotions can increase productivity in all areas including virtual reality, psychology, behavior modeling, in short, human-computer interaction. In this study, we propose a method based on improving the accuracy of emotion recognition using speech data. In this method, new features are determined using convolutional neural networks from MFCC coefficient matrices of speech records in Crema-D dataset. By applying particle swarm optimization to the features obtained, the accuracy was increased by selecting the features that are important for speech emotion classification. In addition, 64 attributes used for each record were reduced to 33 attributes. In the test results, 62.86% accuracy was obtained with CNN, 63.93% accuracy with SVM and 66.01% accuracy with CNN+BPSO+SVM.

Facial recognition is used efficiently in human-computer interactions, passports, driver’s licence, border controls, video surveillance and criminal identification, and is an important biometric’s security option in many device-related security requirements. In this paper, we use Eigenface recognition based on the Principal Component Analysis (PCA) to develop the project. PCA aims to reduce the size of large image matrices and is used for feature extraction. Then, we use the euclidean distance method for classification. The dataset used in this project was obtained by AT&T Laboratories at Cambridge University [1]. The training dataset contains grayscale facial images of 40 people; each person has 10 different facial images taken from different angles and emotions. This study aims to give researchers a hunch before they start to develop image recognition using deep learning methods. It also shows that face recognition can be done without deep learning.

It is an active field of study in studies where the iris center is referenced, such as iris center detection, gaze tracking, driver fatigue detection. In this study, an approach for real-time detection of iris centers based on convolutional neural networks is presented. The GI4E dataset was used as the dataset for the proposed approach. Experimental results estimated the test data of the proposed convolutional neural network model with an accuracy of 97.2% based on the 0.025 error corresponding to the closest position to the iris center according to the maximum normalized error criteria. The study was also tested in real time with a webcam built into the computer. While the test accuracy is satisfactory, real-time speed performance needs to be improved.

Solving optimization problems is still a big challenge in the area of optimization algorithms. Many proposed algorithms in the literature don’t consider the relations between the variables of the nature of the problem. However, a recently published algorithm, called “Bayesian Multiploid Genetic Algorithm” exploits the relations between the variables and then solves the given problem. It also uses more than one genotype unlike the simple Genetic Algorithm (GA) and it acts like an implicit memory in order to remember the old but good solutions. In this work, the well-known Multidimensional Knapsack Problem (MKP) is solved by the Bayesian Multiploid Genetic Algorithm. And the results show that exploiting relations between the variables gets a huge advantage in solving the given problem.

Text mining, which derives information from written sources such as websites, books, e-mails, articles, and online news, processes and structures data using advanced approaches. The vast majority of SMS (Short Message Service) messages are unwanted short text documents. Effectively classifying these documents will aid in the detection of spam. The study attempted to identify the most effective techniques on SMS data at each stage of text mining. Four of the most well-known feature selection approaches were used, each of which is one of these parameters. As a result, the strategy that yielded the best results was chosen. In addition, another parameter that produces the best results with this approach, the classifier, was determined. The DFS feature selection approach produced the best results with the SVM classifier, according to the experimental results. This study establishes a general framework for future research in this area that will employ text mining techniques.

Hodgkin-type lymphoma is a disease with unique histological, immunophenotypic, and clinical features. This disease occurs in nearly 30% of all lymphomas. Its treatable is high. However, the treatment plan is specified after the stage and risk status are determined. For this reason, it is an important process for doctors to decide on the stage of the disease correctly. Some of the data used for this decision are the patient's history, detailed physical examination, laboratory findings, imaging methods and bone marrow biopsy results. Hybrid FDG-PET is the other method used in the medical world. This method is used in diagnosis, evaluation of response given to treatment, staging and restaging process. However, it is radiation-based. Therefore it has the possibility of producing undesirable results in the future. In this study, an artificial intelligence-based computer-assisted decision support system is done to reduce the number of used medical methods and radiation exposure. Data were obtained from the NCBI-GEO dataset. The evaluation of these data, which contains missing values, is handled in two ways. Firstly, samples with missing values in the initial evaluation are deleted from the dataset. Then, these data are trained with “trainlm” function in artificial neural network architecture. However, reducing the error value of the estimates is important. For this, the artificial neural network architecture is retrained with the artificial bee colony algorithm, particle swarm optimization algorithm and invasive weed algorithm, respectively. Secondly, the same operations are performed again on the dataset containing missing values. As a result of the training, the maximum performance was obtained for invasive weed and particle swarm optimization algorithms with 1,45547E+14 and 1,23103E+14 average error rates, respectively.

COVID-19'un tüm dünyada hızla yayılması, birçok ülkenin eğitim kurumlarını geçici olarak kapatma ve uzaktan eğitime geçme kararı almasına neden olmuştur. Zaman ve mekan kısıtlaması olmaksızın uygulanabilen uzaktan eğitim modeli bu süreçte eğitimin sürdürülebilirliği açısından faydalı olmuştur. Ancak, herhangi bir ön hazırlık yapılmadan uzaktan eğitime ani geçiş yapılması beraberinde birtakım sorunları da ortaya çıkarmıştır. Bu bağlamda, uzaktan eğitim süreçlerine katkı sağlaması açısından bu süreçte yaşanan sorunların tespit edilmesi önem arz etmektedir. Bu çalışmanın amacı, COVID-19 pandemisi döneminde öğrencilerin uzaktan eğitimde yaşadıkları sorunları güncel metin madenciliği yöntemleri ile tespit etmektir. Bu amaçla öğrencilerin uzaktan eğitim sürecindeki destek kayıtlarına ait metinler toplanarak bir veri havuzu oluşturulmuş ve bu veriler metin madenciliği yöntemlerinden olan Kelime Bulutu ve Gizli Anlamsal Dizinleme (GAD) teknikleri ile analiz edilmiştir. Analiz sonucunda, uzaktan eğitim sistem altyapısından kaynaklandığı anlaşılan sorunlar (derslerin sistemde görülmemesi, açılmaması, ses sorunları), internet, sınav ve ödev ile ilgili sorunlar, en çok öne çıkan sorunlar olarak tespit edilmiştir., The rapid spread of COVID-19 all over the world has caused many countries to decide to temporarily close their educational institutions and switch to distance education. The distance education model, which can be applied without time and space restrictions, has been beneficial for the sustainability of education in this process. However, the sudden transition to distance education without any preliminary preparation has also revealed a number of problems. In this context, it is important to identify the problems experienced in this process in order to contribute to distance education processes. The aim of this study is to identify the problems experienced by students in distance education during the COVID-19 pandemic with current text mining methods. For this purpose, a data pool was created by collecting the texts of students' support records in the distance education process and these data were analyzed with Word Cloud and Latent Semantic Indexing (LSI) techniques, which are text mining methods. As a result of the analysis, the problems that were understood to be caused by the distance education system infrastructure (courses not being seen or opened in the system, audio problems), internet, exam and homework problems were identified as the most prominent problems.

Bir yazılım geliştirme projesi üzerinde çalışan grup üyelerinin kişilikleri, ekip üyelerinin rollerine uygun olmasına bağlı olarak, çalışma sonucunun başarısını etkiler. Bu nedenle, ekip üyelerinin yazılım projelerindeki rollerini kişilik farklılıklarına göre belirlemek önemlidir. Bu çalışma, yazılım projelerinde yazılım mühendislerinin uygun rollerini belirlemek ve yazılım sektöründeki yetenek yönetimi için, yenilikçi ve yaratıcı bir uzman sistem önermektedir. Önerilen sisteminin en belirgin yönü, teste girenler için kişilik analizi hissini uyandırmayan sorulardan oluşmasıdır. Sorular, yazılımcıların projelerindeki görev ve doğal davranışlarıyla alakalı olacak şekilde tasarlanmıştır. Yazılım mühendisleri genellikle psikometrik testlere inanmazlar ve bunları manipüle etmeye çalışırlar. Ek olarak, bu testlerdeki soru sayısı oldukça fazla olduğundan sıkılabilmektedirler. Önerilen test, Jung Tabanlı Yazılım Yetenek Yönetimi (JBSTM) ve Keirsey Mizaç Sıralayıcı testinin sonuçları ile karşılaştırılmıştır. Bu iki test, 20 yazılım sektörü çalışanına rastgele bir sıra ile tek bir oturumda uygulanmıştır. Testlerden sonra da bir karşılaştırma anketi doldurtulmuştur. E / I, T / F ve J / P'de JBSTM ve Keirsey testi arasında orta derecede bir korelasyon ortaya çıkmıştır. JBTSM ve Keirsey test sonuçları test tamamlanma süreleri açısından da karşılaştırılmıştır. Karşılaştırma anketinin sonuçları kullanılarak JBTSM ve Keirsey testi kolaylık, uygunluk, rahatlık ve kullanışlılık açısından da karşılaştırılmış ve JBTSM’in dört açıdan da Keirsey testine kıyasla daha yüksek puanlar aldığı görülmüştür., The personalities of the group members working on a software development project affect the success of the work result, depending on the suitability of the team members for their roles. Therefore, it is important to determine the roles of team members in software projects according to their personality differences. This study proposes an innovative and creative expert system for determining the appropriate roles of software engineers in software projects and talent management in the software industry. The most obvious aspect of the proposed system is that it consists of questions that do not evoke the feeling of personality analysis for those who take the test. The questions are designed to be relevant to the tasks and natural behaviors of the software engineers in their projects. Software engineers often do not believe in psychometric tests and try to manipulate them. Furthermore, the number of questions in these tests is quite large, so they can get bored. The proposed test was compared with the results of the Jung-Based Software Talent Management (JBSTM) and the Keirsey Temperament Sorter test. These two tests were applied to 20 software industry employees in a single session in a random order. A comparison questionnaire was completed after the tests. A moderate correlation between JBSTM and the Keirsey test appeared in E / I, T / F and J / P. The JBTSM and Keirsey test results were also compared in terms of test completion times. Using the results of the comparison questionnaire, the JBTSM and Keirsey test were also compared in terms of easiness, suitability, comfort and usefulness, and it was found that the JBTSM received higher scores compared to the Keirsey test in all four respects.

Most transportation mode (i.e., train, car) identification methods can only recognize the activities that were previously seen in the training data. However, they cannot be able to detect an unseen activity without having any corresponding training sample. In this study, we propose a k-shot learning algorithm. When k is set to zero, named zero-shot learning, it can recognize a previously unseen new transportation mode (i.e., bus) even when there are no training samples of that mode in the dataset. The experiments carried out on a real-world dataset showed that the accuracy rates from 89.46% to 93.94% were achieved by the proposed method with different values of parameter k. The results also showed that our method outperformed the state-of-the-art methods in terms of classification accuracy.

Optical burst switching networks are vulnerable to various threats including Burst Header Packet Flooding attack, Circulating Burst Header attack, Address Spoofing, and Replay attack. Therefore, detecting such threats play a key role in taking appropriate security measures. One of the major challenges in identifying the risks of Burst Header Packet flooding attacks is the lack or insufficiency of reliable historical data. In this paper, firstly, Burst Header Packet flooding attacks are classified into four categories, Misbehaving-Block, Behaving-No Block, Misbehaving-No Block and Misbehaving-Wait, using Naive Bayes and K-Nearest Neighbor algorithms. Using performance metrics obtained both after testing on the same set and after applying 10-fold cross validation, the performance of Naive Bayes and K-Nearest Neighbor algorithms is compared based on commonly used performance metrics. As the results show, compared to Naive Bayes, K-Nearest Neighbor algorithm is more suitable for predicting Burst Header Packet Flooding attacks.

Stock price prediction is an important topic for investors and companies. The increasing effect of machine learning methods in every field also applies to stock forecasting. In this study, it is aimed to predict the future prices of the stocks of companies in different sectors traded on the Borsa Istanbul (BIST) 30 Index. For the study, the data of two companies selected as examples from each of the holding, white goods, petrochemical, iron and steel, transportation and communication sectors were analyzed. In the study, in addition to the share analysis of the sectors, the price prediction performances of the machine learning algorithm on a sectoral basis were examined. For these tests, XGBoost, Support Vector Machines (SVM), K-nearest neighbors (KNN) and Random Forest (RF) algorithms were used. The obtained results were analyzed with mean absolute error (MAE), mean absolute percent error (MAPE), mean squared error (MSE), and R2 correlation metrics. The best estimations on a sectoral basis were made for companies in the Iron and Steel and Petroleum field. One of the most important innovations in the study is the examination of the effect of current macro changes on the forecasting model. As an example, the effect of the changes in the Central Bank Governors, which took place three times in the 5-year period, on the forecast was investigated. The results showed that the unpredictable effects on the policies after the change of Governors also negatively affected the forecast performance

Bu makalenin amacı dikkat yanlılığı süreçleri ve bu esnada oluşan göz hareketleri üzerine yapılan araştırmaları inceleyerek, bireylerin olumlu-olumsuz uyaranlardan ne şekilde etkilendiğini değerlendirmektir. Bireylerin çevresel uyaranlara karşı gösterdikleri seçici dikkat süreçleri ve buna bağlı geliştirdikleri göz hareketi davranışları, hayatta kalma sırasında oluşan otonom savunma mekanizmaları ile ilişkilendirilmektedir. Bu durum bireyin sahip olduğu çeşitli özelliklere göre farklılık göstermektedir. Farklılık gösteren bir diğer durum ise dikkat değişim süreci içerisindeki bireyin psikolojik iyilik halidir. Birçok çalışma kaygı ilişkili veya depresyon gibi hastalıklara sahip bireylerin tehdit içerikli uyaranlara karşı daha yoğun hassasiyet gösterdiklerini belirtmiştir. Göz takip yöntemi, bireyin dikkat değişim sürecini ve dikkat yanlılığına yansıyan bireysel farklılıkların ayırt ve tespit edilmesi açısından önem arz etmektedir. Bu yöntem ile yapılan çalışmalar sonucunda, kaygı/depresyon seviyesi yüksek bireylerin negatif uyarana gösterdiği tepki süresinin diğer (pozitif, nötr) uyaranlara kıyasla gösterdiği farklılıklar dikkat çekmektedir. Tehdit edici unsurların bireyde yarattığı dikkat yanlılığı süreçleri günümüzde incelenmeye devam etmektedir. Bu duruma ilişkin bilgiler birçok psikolojik hastalığın tespit ve tedavi süreçleri için oldukça önemlidir. Ayrıca göz hareketi davranışının gözlemlenmesi, dikkat yanlılığına neden olan diğer psiko-fizyolojik durumların (bilişsel, görsel yorgunluk) tespiti için de önem arz etmektedir., The purpuse of this review paper was to examined the study on visual attentional bias processes and eye movements in this process and evaluate individuals' responses to positive and negative stimulus. The selective attention processes that individuals show to environmental stimuli and the eye movement behaviors. Accordingly, these are associated with autonomous defense mechanisms that occur during survival. This situation may differ according to various characteristics of the individual. Another differentiated situation is the psychological well-being of the individual during the attention change process. Many studies have shown that individuals with anxiety-related disorders or depression show more intense sensitivity to threatening stimuli. The eye-tracking method is important for characterizing and detecting individual differences reflected in the individual's attention switching process and attentional bias. Consequently, the differences in the reaction time of individuals with high levels of anxiety/depression were negative stimuli compared to other (positive, neutral) stimuli. Currently, the processes of attentional bias caused by threatening factors in individuals continue to be examined. Information on this situation are quite important for the detection and treatment processes of many psychological diseases. Therefore, the observation of eye movement behaviour that is significant for the detection of other psycho-physiological conditions (cognitive, visual fatigue) that cause attentional bias.

Bu çalışmada, dünya ticaretinin en önemli yapı taşını oluşturan özellikle deniz ticaret yollarının ve deniz ulaşımının yıllarca başının en büyük derdi olmuş; ölüm, yaralanma ve bilumum kayıplara sebep olmuş olan deniz kazası/olaylarının başlıca nedenlerine ışık tutularak incelenmesi amaçlanmıştır. Bu bağlamda söz konusu deniz kazaları/olayları sınıflandırılarak hangi çeşit deniz kazasının hangi çeşit sebeplerden meydana gelebildiği araştırılacaktır. Çalışmada, geçmiş yıllara ait deniz kazalarına ilişkin bilgilerden oluşan bir veri setinin, makine öğrenmesi yöntemleri ile eğitilen bir model yardımıyla gelecekte oluşabilecek deniz kazası olasılıklarının tahmin edilmesi amaçlanmıştır. Söz konusu makine öğrenme modeli, çeşitli Gözetimli Makine Öğrenme algoritmaları ile Yapay Sinir Ağları (Artificial Neural Network- ANN) konularını esas alarak Python yazılım dili aracılığıyla geliştirilecektir. Bu çalışmada ulaşılmak istenen esas amaç deniz kazalarına ilişkin önceden tahmin yapılabilecek bir sistemin geliştirilerek, bu kazaların önlenmesi veya bu acı tecrübelerin yaşanmadan önce tedbir alınmasına yönelik bir farkındalık oluşturmaktır., In this study, maritime trade routes and maritime transportation, which constitute the most important building block of world trade, have been the biggest problem for years; It is aimed to examine the main causes of maritime accident/incidents that have caused death, injury and all kinds of losses by shedding light on. In this context, by classifying the said maritime accidents/incidents, it will be investigated which types of maritime accidents may occur due to which reasons. In the study, it is aimed to predict the probabilities of sea accidents that may occur in the future with the help of a data set consisting of information on previous years' maritime accidents, with the help of a model trained with machine learning methods. The machine learning model in question will be developed through the ptyhon software language, based on various Supervised Machine Learning algorithms and Artificial Neural Networks (ANN). The main aim to be achieved in this study is to develop a system that can predict maritime accidents and to raise awareness about preventing these accidents or taking precautions before these painful experiences occur.

Yapay Sinir Ağları (YSA), matematik ve mühendislik problemlerinin çözümünde sıkça kullanılmaktadır. YSA, canlıların beyin sinir hücresinden esinlenerek ortaya atılan ve bir ya da birden fazla nöronun belirli disiplin çerçevesinde bir görevi yerine getirmesini sağlayan matematiksel modeldir. YSA’ların eğitimi aşamasında probleme göre değişebilen gizli katman sayısı, ağırlıklar, öğrenme katsayısı ve daha birçok parametre kullanılmaktadır. Bu çalışmada, Karcı Sinir Ağı (Karcı Neural Network -Karcı NN) olarak adlandırılan ve YSA’nın öğrenmesi aşamasında kullanılan öğrenme katsayısının kullanımı yerine kesir dereceli türev kullanılan yeni bir hesaplama yöntemi kullanılmıştır. Karcı NN yöntemiyle yapılan deneysel çalışmalar sonucunda, özellikle alfa parametresinin 1.4 değeri için hata oranının % 0.019 olarak ölçüldüğü başarılı sonuçlar elde edilmiştir., Artificial Neural Networks (ANNs) are frequently used in solving mathematical and engineering problems. ANN is a mathematical model inspired by the brain nerve cell of living things and enables one or more neurons to perform a task within the framework of a specific discipline. During the training of ANNs, the number of hidden layers, weights, learning coefficients, and many other parameters that can change according to the problem are used. In this study, a new calculation method called Karcı Neural Network (Karcı NN) and which uses fractional derivatives instead of the learning coefficient used in the learning phase of ANN, was used. As a result of the experimental studies carried out with the Karcı NN method, successful results were obtained, especially for the 1.4 value of the alpha parameter, where the error rate was measured as 0.019%.

Finansal zaman serisi tahmini ile finansal varlıklar için doğru alım-satım kararları vererek karlılığın arttırılması amaçlanmaktadır. Finansal varlıkların fiyatları pek çok faktörden etkilenen kırılgan bir yapıdadır. Bu nedenle, finansal zaman serisi tahmini uzun yıllardır farklı disiplinlerden araştırmacılar tarafından ilgi gören oldukça zorlu bir görevdir. Bu çalışmada, günlük ons altın fiyat yönünün tahmini için 2007 – 2021 yıllarını kapsayan 15 yıllık tarihsel fiyat verisi kullanılmıştır. Altın fiyat verileri mum grafikleri ve teknik analiz göstergeleri yardımıyla grafik görüntülere dönüştürülmüştür. Bu sayede altın fiyat yön tahmini 2-sınıflı görüntü sınıflandırma problemine indirgenmiştir. Görüntülerin sınıflandırılması için öncü ön-eğitimli evrişimsel sinir ağı modellerinden AlexNet ince-ayarlanarak adapte edilmiştir. Gerçekleştirilen deney sonuçlarına göre, önerilen yaklaşımın sınıflandırma performansı doğruluk, duyarlılık, hassasiyet ve f-ölçütü performans metrikleri için sırasıyla %53,8, %66,97, %37,54 ve %42,05 olarak ölçülmüştür. Ayrıca önerilen yaklaşımın tahminlerine dayalı gerçekleştirilen ticaret stratejisinin karlılık analizleri de yapılmış ve yatırımcılar tarafından sıklıkla kullanılan Göreceli Güç Endeksi ve Al ve Tut yatırım stratejileri ile karşılaştırılmıştır. 3 yıllık vade boyunca gerçekleştirilen piyasa benzetim sonuçlarına göre önerilen yaklaşım %51,77 kar oranıyla diğer yatırım stratejilerinden daha iyi sonuç vermiştir., With the financial time series forecasting, it is aimed to increase the profitability by making the right buying-selling decisions for financial assets. The prices of financial assets are in a fragile structure affected by many factors. Therefore, financial time series forecasting is a very challenging task that has attracted attention by researchers from different disciplines for many years. In this study, 15 years of historical price data covering the years 2007 – 2021 are used to predict the daily ounce gold price direction. Gold price data has been converted into graphic images with the help of candlestick charts and technical analysis indicators. In this way, gold price direction prediction is reduced to a 2-class image classification problem. AlexNet, one of the pioneering pre-trained convolutional neural network models for the classification of images, was fine-tuned and adapted. According to the experimental results, the classification performance of the proposed approach was measured as 53.8%, 66.97%, 37.54% and 42.05% for accuracy, sensitivity, specificity and f-score performance metrics, respectively. In addition, profitability analyzes of the trading strategy based on the predictions of the proposed approach were also made and compared with the Relative Strength Index and Buy and Hold investment strategies, which are frequently used by investors. According to the market simulation results carried out during the 3-year maturity, the proposed approach yielded better results than other investment strategies with a profit rate of 51,77%.

Demand forecasting is of crucial importance for make-to-stock companies because product demand is uncertain and it changes with time. Fuzzy linear programming (FLP) can be an optimum approach for such uncertain situations. In this study, the FLP model was used for more accurate demand forecasting in a make-to-stock company. Demand forecasting study was carried out according to the FLP method and linear programming (LP) method. Solutions of FLP and LP were compared in terms of imputed shortage cost, inventory carrying cost, and net profit. Results show that the applied FLP method is more advantageous than LP as it provides a 67% decrease in costs and a 15% increase in net profit.

In today's world, the consumption of clothing and accessories is increasing day by day. In this consumption area, people sometimes have difficulty in finding a color outfit or jewelry they want. Especially in face-to-face shopping, people look at all the stores to find the color they are looking for and consume a lot of energy in those stores. Considering this increasing problem, a model is proposed in this study that detects color and then displays the images of the clothing and accessory data it contains according to the color it perceives. In this way, this model provides ease of finding what people are looking for. Therefore, it could be easily stated that this system turns the filtering process into a kind of visual filtering. The results derived using the proposed model prove its effectiveness for detecting the colors.

Yapay zekâ kavramı 1900’lü yılların ikinci çeyreğinden itibaren hayatımıza girmeye başlamıştır. Bugüne kadar ki süreçte bazen çok popüler olmuş, bazen de unutulmaya yüz tutmuştur. 2000’li yıllarda bilgisayar donanımlarının gelişmesi ve yapay sinir ağlarındaki gelişmeler Yapay zekâ araçlarının tekrardan yoğun kullanılmasına sebep olmuştur. Derin öğrenme çalışmaları yapay zekâ çalışmalarının lokomotifi konumundadır. Derin öğrenme sadece bilgisayar biliminde ve nesne algılama görevinde değil, birçok disiplinde ve birçok görevde kullanılmaktadır. Derin öğrenme çalışmaları için farklı mimariler, farklı donanımlar, farklı yazılım çerçeveleri geliştirilmiştir. Derin öğrenme araçlarının ve mimarilerinin bu kadar fazla olması özellikle problemlerinin çözümü için derin öğrenme araçlarını kullanmak isteyen araştırmacıların kafasını karıştırabilmekte veya işini zorlaştırabilmektedir. Bu çalışmada derin öğrenme kavramı hakkında bir anlayış oluşturmayı hedefliyoruz. Derin öğrenme kavramını ve onu oluşturan alt bileşenleri sistematik olarak sunuyoruz. Akabinde günümüzde yaygın olarak kullanılan ana akım derin öğrenme çerçevelerini sunuyoruz. Bu çerçevelerin performans, zaman, doğruluk, Google da aranma, GitHub’da takip edilme gibi farklı ölçütlere göre incelendiği araştırmalardan bir derleme hazırlanmıştır. Bu çalışmanın özellikle derin öğrenme araçlarına aşina olmayan ancak çalışmalarında derin öğrenme araçlarını kullanmak isteyen araştırmacı ve okuyucular için bir kılavuz olmasını arzu ediyoruz., The idea of artificial intelligence has been in our lives since the second quarter of the 20th century. Up to now, it has enjoyed great popularity in some cases, and in others it has been forgotten. In the 2000s, the development of computer hardware and developments in artificial neural networks led to an extensive use of artificial intelligence algorithms. Deep learning studies are the pioneers of artificial intelligence studies. Deep learning is not only used in computer science and object detection, but also in many disciplines and many tasks. Different architectures, hardware and software frameworks have been developed for deep learning studies. The fact that there are so many deep learning tools and architectures can confuse the work of researchers who want to use deep learning tools to solve their problems. In this study we want to create an understanding of the concept of deep learning. We present the concept of deep learning and its subcomponents systematically. We then introduce the common deep learning frameworks that are widely used recently. A review was created from the studies in which these frameworks were examined according to different criteria such as performance, time, accuracy, searched on Google, followed by GitHub. In this study, we would like to be a guide for researchers and readers who are unfamiliar with deep learning algorithms but want to use deep learning tools in their studies.