329
Archived Theses
0
DOIs Assigned
0%
DOI Rate
Archived Theses
Dijital sürdürülebilirlik perspektifinden tekstil üretiminde bir BT yatırım değerlendirme çerçevesi
The textile industry, one of Türkiye's and the world's leading and most significant industries, is undergoing an intensive digitalization process driven by the need to enhance competitiveness through operational efficiency, social compliance, and sustainability requirements. This paper presents a comprehensive assessment framework to evaluate IT (Information Technologies) investments in textile production from a digital sustainability perspective. The framework structured around eight key criteria: technological infrastructure, cost, data security, strategic fit, organizational culture, management support, environmental impact, incentives and policies. Each main criterion is further broken down into sub-criteria to facilitate a systematic and detailed analysis. The evaluation applied on nine IT investment alternatives: Internet of Things (IoT), Digital Twin(DT), Enterprise Resource Planning ERP) Systems, Image Processing (IP), Energy Management Systems (EMS), Manufacturing Execution Systems (MES), Smart Logistics and Inventory Management (SL/IM), Block-chain (BC), and Augmented Reality and Virtual Reality Applications (AR/VR) across eight criteria and twenty-two sub-criteria. To address the uncertainties inherent in decision-making processes, Spherical Fuzzy Logic integrated with AHP and TOPSIS are employed. This hybrid approach combines expert opinions and industry insights, providing a robust and flexible prioritization process under complex, real-world conditions. Preliminary findings indicate that the proposed framework not only helps decision- makers in selecting the most suitable IT solutions for digital transformation but also supports sustainable development in the textile industry by balancing economic, environmental and social objectives. Although the primary focus is on Türkiye's textile industry, the adaptable nature of the framework supports its application in a variety of industries and international environments. Future research directions and potential limitations of the study are discussed, paving the way for further improvements and empirical validation.
Airbnb fiyatlarının, konum ve fotoğraf verileri kullanarak tahminlenmesi
Until recently, temporary accommodation needs were met through businesses such as hotels. However, in recent years, housing rentals through Airbnb have become increasingly common. This model has several distinguishing features. First, the physical characteristics of rental properties (such as the number of rooms and available amenities) show significant variations, both compared to traditional accommodations and among themselves. Second, properties available for rent via Airbnb can be found in almost any location. Finally, the services offered vary compared to hotels. These differences necessitate different approaches to predicting prices. Identifying the factors that influence pricing accurately is crucial for profitability. This study, based on Airbnb listings in Istanbul, analyzes which factors influence property prices and to what extent, as well as determining the most successful price prediction method. In the research, in addition to basic information commonly included in Airbnb listings, factors like the integration of the location with public transportation, the extent to which daily needs can be met, and the analysis of listing photos were also considered. Various machine learning techniques were applied for price prediction, and the impact of each factor on pricing was calculated. The results indicate that while the physical characteristics of a property play a significant role, location-based information is also crucial. Proximity to public transportation stops, cultural activities, and educational facilities were identified as key factors. It was also determined that analyzing the listing photo using a large language model did not provide a meaningful contribution to price prediction.
Web erişilebilirliği değerlendirmesine PA11Y tabanlı bir yaklaşım
In today's world, the impact of digital technologies on social life is increasing, and access to information is predominantly facilitated through digital platforms. This development necessitates that publicly available websites be accessible to all individuals. Ensuring equal digital rights for people with disabilities is not only an ethical responsibility but also a legal obligation. Website accessibility evaluation processes are generally carried out using automated testing tools. However, existing tools have significant limitations, particularly in analyzing dynamically generated content. They often produce false positives and fail to provide comprehensive assessments. In this thesis, an accessible evaluation system has been developed based on the open-source tool Pa11y, offering Turkish language support and visual feedback mechanisms. The system features a Node.js-based backend and a React.js-based frontend, enabling users to visually detect and interactively correct accessibility issues such as insufficient color contrast, missing alternative texts, and unlabelled form elements. With proxy support, the system overcomes CORS restrictions and can analyze both static and dynamic content. Furthermore, it includes features such as saving analysis history, performing comparative evaluations, and exporting reports in PDF/JSON formats. Applied tests and user feedback indicate that the system produces more accurate and consistent results compared to existing tools. In particular, it successfully analyzes dynamic content that popular tools like WAVE fail to process properly. The system's Turkish localization and visual feedback features significantly enhance user experience. This study represents a concrete contribution to improving the accessibility of digital services in Turkey and lays the groundwork for future developments in compliance with WCAG 2.2 and beyond.
İşbirlikçi filtreleme performansının çok boyutlu analizi: Metrikler, topolojiler ve algoritmalar
Recommendation systems play a crucial role in digital platforms by offering relevant items to users based on their interaction histories. Although considerable progress has been made in developing new algorithms, evaluating these systems remains challenging due to the overreliance on accuracy metrics and limited understanding of how dataset characteristics influence model performance. This thesis proposes a comprehensive evaluation framework for personalized collaborative filtering models (PCFMs) through a systematic analysis of 13 recommendation algorithms across 10 real-world bipartite datasets, using 12 performance metrics spanning accuracy, diversity, and fairness. Our experimental design is structured around four research questions, examining performance variations across evaluation dimensions, the coherence among metrics, model sensitivity to metric types, and the impact of graph topological features on performance. The results show that graph-based models achieve more balanced outcomes across all three evaluation aspects. Additionally, while accuracy metrics exhibit strong internal correlations, diversity and fairness metrics display more complex and model-dependent relationships. Moreover, topological features such as sparsity and L3 similarity significantly affect performance, though their influence varies by metric and model type. Overall, this study demonstrates that reliable evaluation of recommendation systems requires a multidimensional perspective, offering insights that are essential for both academic benchmarking and real-world deployment.
Elektrik sinyallerinden makine öğrenmesi ile cihaz kategorizasyonu
The growing need for energy management and sustainability necessitates innovative solutions in the analysis of electrical signals and device recognition. This study presents a novel approach for categorizing devices based on their unique energy consumption patterns using electrical signals. A TinyML-based system has been developed to enable resource-efficient and scalable device recognition on embedded platforms. The research addresses critical challenges in device recognition systems, such as signal noise, overlapping energy profiles, and real-time processing constraints. By leveraging advanced analytical techniques, the proposed system improves the accuracy and efficiency of device categorization. Additionally, it contributes to the broader adoption of TinyML in energy monitoring systems, providing significant benefits in terms of energy efficiency, fault detection, and sustainability. This study bridges the gap between academic innovation and practical application, supporting the development of smart home technologies, industrial automation, and intelligent energy management systems. The findings demonstrate the potential of integrating TinyML into energy monitoring frameworks, paving the way for more sustainable and user-friendly solutions.
Kentsel dünyanın 3D algısı için derin öğrenme tabanlı tespit ve segmentasyon
The rapid development of autonomous driving technologies has highlighted the growing need for perception systems that can understand and interpret complex urban traffic scenes. This thesis presents a deep learning-based, vision-only framework that focuses on the detection and segmentation of three key elements in traffic environments: roads, vehicles, and pedestrians. The system operates using 2D image projections generated from 360-degree cameras, without relying on LiDAR or any other external sensors. This approach aims to offer a scalable and cost-effective solution for intelligent transportation applications. To train and evaluate the proposed models, a custom dataset was created using Google Street View imagery collected from four major European cities: Istanbul, Paris, Munich, and Marseille. The dataset includes 8,932 labeled images and more than 149,000 object annotations, providing a diverse range of traffic scenes under varying urban conditions. YOLOv8 and YOLOv10 models were used for object detection, while DeepLabV3 was applied for semantic segmentation. The models were evaluated across different train-validation splits using standard metrics. YOLOv10 achieved the best detection performance with a mAP@0.5 score of 0.685 and an overall precision of 0.76. DeepLabV3 produced strong segmentation results, including IoU scores above 0.85 for road detection and F1 scores exceeding 0.80 in clearly defined object regions. These results indicate that image-only systems can effectively perform traffic scene analysis in real time. The framework developed in this thesis demonstrates the potential of deep learning for urban scene understanding and contributes a new dataset that supports future research in vision-based autonomous navigation.
Afet yönetiminde tahmin ve optimizasyon uygulamaları
This study considers the preparedness and response phases of a disaster management problem. One of the biggest problems for disaster management is that accurate predictions for the number of death and injured people are not available. Researchers in disaster management usually use data from the historical disasters as if they were the true numbers of death and injured people. However, in a future disaster, the numbers of injured and death people will be different with probability one. Consequently, recent studies focus on predicting these numbers applying different machine learning techniques such as neural networks, support vector machine, support vector regression, Gaussian process, and multiple linear regression. In this study, we use Gaussian processes to predict the numbers of injured people. This prediction method uses only two types of explanatory variables, namely, population density and the number of undamaged, slightly and moderately damaged buildings. We use the data resulting from the two earthquakes occurred on 6 February 2023 in the southeastern region of Turkey, and we collect the data from different sources. We compare our prediction method with three other benchmark prediction methods. None of the methods provide very accurate predictions. However, Gaussian processes have the advantage of estimating the variance of the predictor in addition to the predictor, and the predictor has normal distribution. We further use these predictors and their variances to solve a chance-constrained optimization problem to find shelter locations. We use randomly generated data for the optimization problem. Further research has to be done to provide: i- machine learning method with better prediction capacity; ii- solution of the shelter location problem with real data; iii- solution procedure to solve a large-scale optimization problem for shelter location problem.xii Keywords : Machine learning, Gaussian process, Chance constraints, Shelter location problem, Disaster management
Nesnelerin interneti bazlı uç bulut sistemleri için konteyner taşıma yaklaşımı
The increasing demand for edge computing and the rise of cloud systems have led to the need for efficient resource management and workload distribution across edge and central cloud environments. Containerization has emerged as a popular solution for application deployment and management due to its lightweight and portable nature. However, migrating containers between edge and cloud environments horizontally and vertically poses challenges in terms of network latency, resource constraints, and service availability. This research project aims to develop a container migration approach specifically tailored for edge-cloud systems. The objective is to design and implement a mechanism that enables seamless and efficient container migration between edge devices and cloud infrastructure. By leveraging the advantages of both edge and cloud computing, it will be possible to allocate resources dynamically based on workload demands and optimize the overall system performance. Key aspects of this project include investigating the challenges and requirements of container migration in edge-cloud systems, designing an efficient migration mechanism that considers network latency, resource availability, and service continuity, evaluating the performance and reliability of the migration approach under various scenarios and workloads, analyzing the cost-effectiveness and resource utilization of the proposed solution compared to traditional deployment models, and developing strategies for load balancing, fault tolerance, and workload optimization to enhance the overall system efficiency. This project aims to provide insights into the feasibility and benefits of container migration in edge-cloud systems. The findings will contribute to the development of more efficient resource management strategies and enable the seamless deployment and migration of containers across edge and cloud environments.
Sürdürülebilirlik perspektifinden küresel bulanık TOPSIS metodolojisine dayalı BT proje yatırım değerlendirme çerçevesi
IT project investmen decision is a challenging process in the current technological dynamic environment because of the multiple criteria and requirements that need to be taken into account. The purpose of this paper is to design a suitable decision model to assess the IT projects, with a unique emphasis on the sustainability context. This study uses the MCDM technique with criteria selected and weighted according to sustainability aspects, including economic, environmental, social, and cultural. Consequently, the Spherical Fuzzy TOPSIS approach is applied to prioritize the IT projects based on their correspondence to the sustainability criteria. A theoretical background of Spherical Fuzzy TOPSIS is introduced to evaluate the sustainability effect of different undertakings. Reviewers of this study identified that Optimization on Flight Planning System as the most sustainable IT project among all the projects in this study and closely followed by the Carbon Offset Project. Based on these findings, this study concludes that projects that are most sustainable are those that directly influence the operational efficiency and resource usage within highly technological industries like the aviation industry especially if one considers long term gains such as; fuel conservation, integration, and effective information processing. To sum up, the conclusions give a clear signal that even though the high-cost projects imply a high level of risk, the achievement of sustainability concerns in many organizations' strategic objectives like operating efficiency, environmental sensitivity, and customer experiences would be enormously beneficial. Currently, there is a lack of information about sustainability analysis of IT projects in the existing literature. This study fills that void by considering sustainability issues in the assessment process and urging decision-makers to approach projects from this perspective. The proposed framework enables organizations to evaluate and select IT projects that best align with their sustainability strategies and operational constraints. Unlike previous studies, which often rely on traditional project evaluation criteria, this research introduces novel criteria such as Impact on Customer Experience, Flexibility, and Compatibility with Existing Systems, offering a more holistic approach to IT project evaluation.
Yaygın ortamlar için olay tabanlı dinamik sensör yapılandırılması
In recent years, the use of Internet of Things (IoT) devices has grown rapidly, leading to the proliferation of IoT networks. As IoT adoption expands, the diversity of devices in use has also increased, resulting in significant heterogeneity across various dimensions, such as communication protocols and data transmission formats. While several methods have been proposed to address these challenges, the majority of existing research focuses on urban IoT applications. In contrast, rural environments typically exhibit greater device heterogeneity due to the wide range of environmental and operational conditions. Equipment selection in such settings is influenced not only by harsh weather and potentially hazardous environments, but also by factors such as energy efficiency and communication capabilities. These constraints limit the available options, making heterogeneity unavoidable. In this study, we propose a middleware solution designed for a real-time, heterogeneous IoT weather sensor network. The system delivers collected data to researchers, administrators, and end users. The middleware is carefully developed with attention to both semantic and syntactic differences among data sources and is integrated with an interface that facilitates access for various stakeholders.