LEE-Mühendislik Yönetimi-Yüksek Lisans

Bu koleksiyon için kalıcı URI

http://hdl.handle.net/11527/19404

Gözat

Yazar "Bayyurt, Nizamettin" ile LEE-Mühendislik Yönetimi-Yüksek Lisans'a göz atma
  • Öge
    Enhancing human resource decision making with image-based OSMI data analysis: leveraging PIX2PIX for accurate workplace mental health insights
    (Graduate School, 2023-06-09) Farid, Fariba ; Bayyurt, Nizamettin ; 507201211 ; Engineering Management
    Mental health issues have become increasingly prevalent and severe in today's society, negatively impacting all aspects of individuals' lives. Artificial intelligence (AI) and its subfields, such as Machine Learning (ML) and Deep Learning (DL), have widespread applications in many domains. These invaluable tools have been widely utilized in various companies to identify mental health disorders and their effects on their employees and workplaces. Many studies have used machine learning models to identify factors that contribute to mental disorders. This study, however, takes a new approach by generating predicted images of mental health disorders among the Tech Survey population. This research provides valuable insights into mental health disorders in the technology industry, which can be used by human resource departments and company leaders to support employees and enhance productivity. Ultimately, this can create a mutually beneficial relationship between workers and employers. The Pix2Pix GAN model is used as a novel technique for the Open Sourcing Mental Illness (OSMI) application. The dataset used in this study comprises over 1484 responses collected via Google Forms during 2017, 2018, 2020, and 2021. This approach acquires further valuable insights on how to make decisions for employees who experience mental health disorders by processing and evaluating data. Preprocessing and scaling the data is a crucial step in Knowledge Discovery in Databases (KDD) as it is challenging to analyze raw data. In this study, the data was preprocessed and scaled before being transformed into images where each pixel represents an attribute. This dataset mainly consists of male individuals residing in the United States who are employed in the technology industry. This is in line with the reality that men hold a greater proportion of positions in the tech industry than women, and It is widely believed that mental health disorders have a negative impact on work performance and productivity in this industry. The thesis highlighted the importance of understanding the characteristics of the questions answered in the data. To achieve this, a close examination of the content was conducted, analyzing various attributes. The survey included different types of questions, such as general, workplace-related, and personal. While the dataset had limited data quality information, it was well-documented. It would have been beneficial to have information on the types of tech companies where participants worked to compare attitudes toward mental health across different sectors. The Mental Health Tech Survey provided useful psychometric variables, which were further categorized and discussed in the thesis.In this study, a significant innovation is the transformation of structured data into unstructured data. This involves changing data that is organized in a specific format,a tabular data, into a format that is not pre-defined, an images. The main aim of this conversion is to enhance data accessibility, improve analysis, and make it more useful for decision-making. To achieve this, the questions were divided into general feature questions asked of employees overall, and questions that are particularly valuable for human resources to gain insights about their employees, such as whether they talk about mental disorders during interviews. These questions contained more self-reported answers that even employees may not be aware of as indicating mental illness. The model was then trained to map the input feature picture to the output label picture, generating a new image that corresponds to the input feature picture. As a result,in this case, the input feature picture represents an individual's answers to questions related to mental health disorders, while the output label picture represents the labels that are more relevant in diagnosing mental disorders based on previous studies.The process of selecting questions as labels or features in the context of tech surveys plays a crucial role in recognizing individuals with disorders and understanding mental health. These questions consider various factors that aid in identifying individuals with disorders, such as self-reporting their condition, comfort in discussing it with employers, seeking treatment, and the impact of the disorder on their lives. Additionally, the selected questions provide insights into how mental health is perceived among employees in the tech industry, allowing for a better understanding of attitudes, norms, and challenges in this context. The focus is on a holistic evaluation that considers the overall well-being of individuals, rather than specific disorders. The self-reported nature of the questions is essential as it reveals how individuals identify themselves and their preferences regarding disclosure. This information helps human resource professionals make informed decisions during the hiring process and provide tailored support to optimize employee performance and well-being. The pix2pix GAN model was then evaluated on a dataset of 1458 images using both quantitative metrics, such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Peak Signal-to-noise Ratio (PSNR), Structural Similarity Index Measure (SSIM), Signal-to-reconstruction Error Ratio (SRE), Spectral Angle Mapper (SAM), and Universal Image Quality Index (UIQ), and qualitative inspection. The results showed that the model was capable of generating high-quality images efficiently. In order to determine the best model for GAN training and prevent overfitting, the dataset was split into three sets: train, test, and validation. The optimal epoch for all sets was determined by visually inspecting the images, comparing various quantitative metrics and visualization methods, and the best epoch was found to be epoch 30. The results of this study indicate that the pix2pix GAN model is a beneficial tool for creating high-quality Mental Health Images (MHI), which may have innovative uses in identifying valuable information. By using images instead of lengthy questionnaires, HR employers can quickly obtain the desired information. Furthermore, the findings reveal a significant correlation between individuals who disclose their mental health issues to potential employers during interviews, mental health conditions that impair their job performance, and individuals with a family history of mental illness. These results could support suggested visionary aims for HR departments to help potential employees who believe that their mental health issues may affect their productivity or performance at work. The predictive capability of the pix2pix GAN holds value for human resource professionals as it can potentially aid in identifying or predicting mental health conditions in employees based on available data. This tool can facilitate the decision-making process for HR professionals and enable early intervention or targeted support to enhance employees' mental well-being.
  • Öge
    Rüzgar santrallerinde elektrik depolama optimizasyonu
    (Lisansüstü Eğitim Enstitüsü, 2024-06-11) Yıldırım, Gülistan ; Bayyurt, Nizamettin ; 507211213 ; Mühendislik Yönetimi
    Türkiye geniş kıyı şeridi ve dağlık bölgeleri ile birlikte önemli ölçüde rüzgar enerji potansiyeline sahiptir. Bu sayede ülkemizde enerji ihtiyacının rüzgar gibi yenilenebilir kaynaklardan sağlanabilmesi mümkün olabilmektedir. Yenilenebilir kaynakların elektrik üretim miktarları elektrik talebinden bağımsız olup, meteorolojik olaylara bağımlıdır. Rüzgar elektrik santrallerinde de durum aynı şekildedir. Elektrik talebinin düşük olduğu saatlerde yüksek rüzgar hızı mevcut ise rüzgar elektrik santrali (RES) üretime devam etmektedir. Tersi durum olan, elektrik talebinin yüksek olması durumunda rüzgar hızı düşük ise, talebin daha az bir bölümü rüzgarlardan karşılanabilmektedir. Bu gibi durumlar enerji arzında istikrarsızlık, sistem dengesinin bozulması, enerji dalgalanmalarının oluşması gibi birçok sorunu da beraberinde getirmektedir. Bu çalışmada İzmir' de bulunan bir rüzgar elektrik santralinin depolama sistemi ile entegre olması durumu araştırılmıştır. Depolama sisteminin sağlayacağı sistem kararlılığı, enerji dalgalanmalarının dengelenmesi, sürdürülebilirliğin sağlanması, talebin yüksek olduğu saatlerde arz kesintinin yaşanmaması gibi yaratacağı birçok fayda dışında elektrik piyasasında oluşan dengesizlik ve KÜPST maliyetlerinde yaratmış olduğu iyileştirme araştırılmış ve fiyat arbitrajı konuları irdelenmiştir. Çalışma Haziran 2023 yılına ait veriler ile yapılmıştır. 6 adet türbini bulunan ve yenilenebilir enerji kaynaklarını destekleme mekanizması (YEKDEM)' na tabii bir santraldir. Bu santral için makine öğrenmesi yöntemleri kullanılarak elektrik üretim tahmin modelleri elde edilmiş olup bu tahmin modellerinin performansları karşılaştırılmıştır. En iyi tahmin performansı CatBoost makine öğrenmesi yöntemi ile elde edilmiştir. Çalışmanın ikinci kısmı olan depolama algoritmasının amacı ise piyasadan kaynaklı oluşan dengesizlik maliyeti ve KÜPST' ü minimize etmek ve fiyat arbitrajı sağlamaktır. Depolama entegrasyonu ile birlikte elektrik talebi düşük iken arz fazlası olan elektriğin depolanması, elektrik talebinin arttığı durumda ise sisteme enerji kazandırılması sağlanır ve anlık olarak santralin üretim tahmini ile gerçekleşen üretim değeri arasındaki farktan kaynaklı oluşan dengesizliğin yine anlık olarak dengelenmesi sağlanır. Bu algoritmanın girdilerinden bir diğeri de Piyasa takas fiyatı tahminidir. Bu veri için gerçekleşen fiyat baz alınmıştır. Depolama algoritması ile birlikte dengesizlik ve KÜPST maliyeti %38 azaltılmış olup, fiyat arbitrajı ile de Haziran ayı toplamında 877,169 TL fayda sağlanmıştır.