LEE- Endüstri Mühendisliği Lisansüstü Programı

Bu topluluk için Kalıcı Uri

http://hdl.handle.net/11527/19266

Gözat

Yazar "Beyca, Ömer Faruk" ile LEE- Endüstri Mühendisliği Lisansüstü Programı'a göz atma
  • Öge
    A hybrid deep learning metaheuristic model for diagnosis of diabetic retinopathy
    (Graduate School, 2022-10-17) Gürcan, Ömer Faruk ; Beyca, Ömer Faruk ; 507142109 ; Industrial Engineering
    Diabetes is a disease that results in an increase in blood sugar due to the pancreas not producing enough insulin, insufficient effect of the produced insulin, or ineffective use of insulin. According to the International Diabetes Federation 2021 report, approximately 537 million adults aged between 20 and 79 live with diabetes worldwide. It is estimated that the number of people with diabetes will increase to 643 million in 2030 and 783 million in 2045. Diabetic retinopathy (DR) is an eye condition that can cause vision loss, irrecoverable visual deterioration, and blindness in people with diabetes. Today, it is one of the leading diseases that cause blindness. Anyone with any diabetes can become a DR. In ophthalmology, type 2 diabetes can lead to DR if left untreated for more than five years. Diabetes-related high blood sugar leads to DR. Over time, having too much sugar in the blood damages the retina. The deterioration of this disease in the eye begins when sugar blocks the capillaries leading to the retina, causing fluid leakage or bleeding at a later stage. The eye produces new vessels to compensate for the blocked vessels, but these newly formed vessels often do not work well and can bleed or leak easily. DR can lead to other serious eye conditions. For example, about one in 15 people with diabetes develop diabetic macular edema over time. DR can lead to the formation of abnormal blood vessels in the retina and prevent fluid from leaving the eye. That causes a type of glaucoma. It is crucial for people with diabetes to have a comprehensive eye examination at least once a year. Follow-up of diabetes; factors such as staying physically active, eating a healthy diet, and using medications regularly can stop the damage to the eye and help prevent or delay vision loss. Some risk factors increase the development of DR, such as pregnancy, uncontrolled diabetes, smoking addiction, hypertension, and high cholesterol. In addition to being detected by magnifying the pupil in eye examination, DR is also diagnosed with the help of image processing techniques. It is common to use fundus images obtained by fundus fluorescent angiography to detect DR and other retinal diseases. Nowadays, with the increasing number of patients and the developments in imaging technologies, disease detection from medical images by various methods has increased. Deep learning is one of the methods whose application area has increased exponentially in recent years. Deep learning is a subfield of machine learning; both are a subfield in artificial intelligence. Deep learning methods draw attention with their versatility, high performance, high generalization capacity, and multidisciplinary use. Technological developments such as the collection of large amounts of data, graphics processing units, the development of robust computer infrastructures, and cloud computing support the building and implementation of new models.Increasing the number of images for a particular patient case and high-resolution images increases specialists' workload. Diagnosis of DR manually by an ophthalmologist is an expensive and time-consuming process. It requires experts who have remarkable experience. In addition, the complexity of medical images and the variations between specialists make it difficult for radiologists and physicians to make efficient and accurate diagnoses at any time. Deep learning is promising in providing decision support to clinicians by increasing the accuracy and efficiency of diagnosis and treatment processes of various diseases. Today, in some medical studies, the success levels of expert radiologists have been achieved or exceeded. Convolutional neural networks (CNNs) are the most widely used deep learning networks in image recognition, image/object recognition, or classification studies. A CNN model doesn't need manually designed features for training; it extracts features from data directly while network training on images. The automated feature extraction property and their success make CNNs highly preferred models in computer vision tasks. This study proposes a hybrid model for the automatic diagnosis of DR. A binary classification of DR (referable vs. non-referable DR) is made using a deep CNN model, metaheuristic algorithms, and machine learning algorithms. A public dataset, Messidor-2, is used in experiments. The proposed model has four steps: preprocessing, feature extraction, feature selection, and classification. Firstly, fundus images are pre-processed by resizing images and normalizing pixel values. The inception-v3 model is applied with the transfer learning approach for feature extraction from processed images. Then, classification is made using machine learning algorithms: Extreme Gradient Boosting (XGBoost), Random Forest, Extra Trees, Bagged Decision Trees, Logistic Regression, Support Vector Machines, and Multilayer Perceptron. XGBoost gives maximum accuracy of 91.40%. The best potential features are selected from the extracted features by three metaheuristic algorithms: Particle Swarm Optimization, Simulated Annealing, and Artificial Bee Colony. Selected features are classified with the XGBoost algorithm. The metaheuristics significantly reduced the number of features obtained from each fundus image and increased the classification accuracy. According to the results, the highest accuracy of 93.12% is obtained from the features selected with Particle Swarm Optimization. When the study results are compared with the existing studies in the literature, it has shown that this study is competitive in terms of accuracy performance and obtained low features. On the other hand, the proposed model has some advantages; it has a few pre-processing steps, training number of parameters are considerable low, and model can be trained with a small amount of data. This study is one of the first studies showing that better results can be obtained in DR classification by using deep learning and metaheuristic algorithms together. The proposed model can be used to give another idea for ophthalmologists in diagnosing DR.
  • Öge
    Business card as a bank product and establishment of a new business card tendency model
    (Graduate School, 2023-04-17) Bozkurt, Onur ; Beyca, Ömer Faruk ; 507191121 ; Industrial Engineering
    Specially issued for the commercial needs of SMEs, the business card is equipped with the features of different products that also allow personal use. Within the scope of this product; commercial credit cards, overdraft accounts, and credit products with equal installments or seasonal payments are combined into a single card. However, during the data analysis phase, it will be seen that this product is used by individual customers as well as SMEs. Individual customers referred to here are customers with legal entities called partnership customers. The business card tendency model currently used for the related product, which has an important place for banks, does not work successfully. The success rate of this model is well below the other models used in the bank, and therefore the accuracy of the data and the model is doubted. In order to improve this situation, the existing model will be observed, deficiencies and errors will be examined, and then a new model will be established. In this process, there will be stages of data preparation, model building, analyzing the output, and evaluating the results. In the literature, there are many studies prepared for credit cards and related issues by banks and various institutions. Some of these are artificial intelligence-supported credit card models, neural approaches to credit scoring, and calculation of the default rate of loans and fraud detection with the help of machine learning. This study aims to design an end-to-end business card modeling process in the light of other studies in a similar context, but with modern approaches that are not thought to be included in the literature. In light of the studies in the above-mentioned literature, it can be said that there are quite advanced approaches to credit cards. The first problem here is that security-related models such as default rate and fraud are emphasized instead of sales-oriented models of credit cards. Another problem is that the existing sales-oriented models do not attach the necessary importance to sustainability and are established with a shorter-term profit and success focus. In general, the studies conducted in the banks related to the subject were investigated, and various similar and different deficiencies were observed in these studies. A similar and common mistake is that the process ends when customers purchase a product that they do not already have. Another problem is the use of customers' information, which is likely to be erroneous and whose accuracy is doubtful, instead of market information with higher validity, such as BKM and GIB. Finally, most models put less emphasis on activity and continuity, and focus on customer balances. During the case analysis phase, the business card product, which can be tracked through monthly sales at the bank, will be examined. In order to establish a model for this product, first the target definition will be determined, the necessary data will be collected from the relevant tables, and some of them will be selected by filtering these data. Afterward, this data set will be prepared for the model-building phase by making the necessary manipulations on it, and then the model-building phase will be started. Following the model setup, the results will be examined, the most appropriate option will be considered, and success will be measured. In this study, the number of variables, which was 83 at the beginning, is reduced to 11 during the model-building phase, in accordance with the principle of parsimony. A meaningful result is tried to be obtained by entering these variables into the logistic regression and random forest models. According to the results obtained, the logistic regression model works with 98% accuracy, while the random forest model works with 99% accuracy. In addition, the precision value obtained in the random forest model is higher than that in the logistic regression model. The precision metric shows how many of the values that are estimated as positive are actually positive. For these reasons, it is decided that the model to be used should be random forest. In this way, the detection rate of customers with a target definition of 1 for the business card product will be higher, which will increase the bank's customer portfolio and profitability. It is aimed in this thesis study to eliminate these deficiencies and errors in existing exercises to establish a more beneficial and efficient model for banks. In addition, it is recommended that banks expand their perspectives on which data they will use while establishing the relevant model. At the end of this process, financial institutions will be able to establish healthier models by integrating and using more accurate and consistent market data into their databases, if necessary. This research will provide a successful trend model with meaningful explanatory variables for business card-like products for future works.
  • Öge
    Short term electricity load forecasting with deep learning
    (Graduate School, 2022-02-25) Yazıcı, İbrahim ; Beyca, Ömer Faruk ; 507142119 ; Industrial Engineering
    In this study, STLF is considered for the real-world case application. STLF horizon spans of half-hour-ahead up to several-day-ahaed timesteps. Energy market establishments have been developed by introducing market regulations in Turkey since 2001. After many regulations and transitions from state-run-market to a non-governmental regulated market, Energy Markets Enterprise Corporation (EPİAŞ in Turkish)was established in 2015. In this market, the day-ahead-market, intraday market and balancing market mechanisms play important roles for the electricity system management in Turkey. These mechanisms plays complementary roles for each other. In this market, stakeholders aim to avoid extra costs arose in balancing market where deficient and excessive amounts of electricity are compensated by purchase and sale among stakeholders since the market imposes 3% penalty costs for these deficient and excessive amounts. And this avoidance can be facilitated by efficient forecasting performance. Hence, forecasting task arises as an important tool for decision makers in forecasting. Before transtion to a regulated market by EPİAŞ, predctions are performed mainly weekly or more than one-week-ahead. The error margins for the predictions made were in turn very high and flexible. This flexibility provided the electricty providers in the market to compromise their excessive and deficient amounts easily when compared to the regılated market situations. Flexibility in the prediction error margins enabled the providers to meet their requirements in the market in the ong horizon with less price charge. In the regulated market, the day-ahead market and intraday market mechanism have turned out to be the integral part of the market mechanisms. Sustaining the competition in the market, growing the market share, reducing the operational costs, and penalty costs created by overestimation and underestimation of the load forecasting, tasks of one-hour-ahead forecasting and one-day-ahead forecasting located at the heart of major concerns for the electricity provider firms in the regulated market. The provider firms in turn focused on these tasks to achieve the aforementioned goals, then create business value through performing these tasks. Thus, this study focuses on the major concerns for the providers by deploying deep learning algorithms for a real-world case. In this study, electricty load data which consists of hourly load demands, for 3 years collected between 2015 and 2017 years were utilized. The granularity of the time series data obtained was composed of load values and temperature values as it is used for the regular forecasting task by the provider firm. In the first stage of applications, preliminary data examinations were performed which provides a guide for time series problem handling for both applications of conventional machine learning, and deep learning methods. These data examinations contained data normalization, dummy variable inclusion, autocorrelation identification tasks for each method type. This stage is followed by input set preparation for the methods deployed. We framed our dataset into a supervised learning dataset by shifting values according to the results of autocorrelation identification, that is time lag. Weekly time lag was found the best choice, hence we used this time lag value in our framing. In addition, since neural networks are at the heart of the applications in this study, we used data normalization as zero-mean normalization to facilitate fast convergence and numerical stability for the networks in training and testing. After preliminary data examinations, we conducted comprehensive comparative analyses of the methods. In the first round of the comparative analyses, two deep learning methods and some popular machine learning methosds were compared whether deep learning methods overcome the conventional methods in STLF task. The deep learning methods were in turn found superior to the conventional methods used which the results were validated by statistical significant test. In the second round of the comparative analyses, just deep learning methods were compared. This round of the comparisons was the central theme in this study since the aim was to propose a deep learning method for the real-world case. For this reason, we proposed a new method based on one-dimensional convolutional neural networks, and compared its performance with the other deep learning methods by applying them to the real-world case. As per the results obtained from this round of comparisons, the proposed method proved its efficiency for both one-hour-ahead, and one-day-ahead prediction tasks. This fact was also validated by statistical significance test as well. In brief of this study, there are some level of takeaways from the results of the study. At the organizational level takeaways, intelligent technqiues use especially in energy sector such as deep learning, deep reinforcement learning tools will make contributions to organizations with different levels. Secondly, energy sector is one of the businesses that enormous amount of data is hoarded even hourly. Hence, creating business value by utilizing intelligent systems in their operations will enable short-term, mid-term, and long-term achievements for them when considered big data regime, advents in hardware and software solutions, and developments in artificial intelligence methods especially in neural networks. At the most conceptual level, deep learning methods provide high-performance forecasting engine for the providers for STLF as per the results obtained. Deployment of these type of artificial intelligence method will make them at the front line in the market. At the method-level takeaways, calendar effects have landmark importance in time series modelling for STLF. Rare time issues, and dual calendar effects are another landmark important issues in time series modelling as well. Efficient feature extraction ability of Convolutional Neural Networks (CNN), and auto-capturing long-term relations in long sequences make them a rival for Long-Short Term Memory (LSTM) and Gated Recurrent Unit (GRU) in time series modelling tasks besides the tasks of audio recognition, speech recognition, natural language processing. In addition, the proposed method's exogenous variable inclusion for modelling the time series problems boosts the performance of the method since different level of resolutions are captured by this setting. Hence, this setting can be extended for later method developments of deep learning methods.