9
Archived Theses
0
DOIs Assigned
0%
DOI Rate
Discipline
Prediction of league of legends draft picks for competitive matches
League of Legends is one of the biggest games in the professional e-sports industry, and it has been a data-driven field for a while. The mechanism of hero selections as to predict which characters will be selected or not in each selection is an important issue. In this study, the mechanism of hero selections in the game has been tried to be examined. Since champion selections are more prominent in competitive and professional matches, the selection rankings of all professional matches in 2020 were obtained by web-scraping methods. After that, the average in-game statistics of all champions were obtained from the Oracle Elixir database, and the two data sets were combined. Machine Learning models were used to analyze drafts. In this context, while the performance of the XGB model was higher for the first three selections, the NB model was ahead of the others for the last two selections. In order to increase the explanatory power of these models and to comment on the characters, Shapley values were used. Models were applied to the World Championship Tournament final in 2020, and the selection probabilities of the champions played in those matches were calculated. Depending on this, these models can be applied for different tournaments and specific teams in the same year. It can help teams and players better analyze their opponents and their own choices.
Using peer to peer communication characteristics to improve gender prediction in online discourse
Using peer to peer communication characteristics to improve gender prediction in online discourse
An analysis of The Central Bank of The Republic of Türkiye (CBRT)'s monetary policy communication with text mining
In the last twenty years, most of the central banks have started to use short term interest rates as a main monetary policy tool to control inflation. Moreover, communication strategy has become one of their most important policy tools through time. Especially, after the 2008 Global Financial Crisis, when the nominal interest rates hit the zero lower bound, central banks have used communication policy very effectively to form expectations. As a result of these developments, there is a growing body of literature examining the communication strategies of central banks. Advances in text mining literature also enables to analyze central bank texts by using alternative methods. Under this background, this thesis analyses the Central Bank of the Republic of Türkiye's (CBRT) Monetary Policy Committee decision announcements for the period 2006-2020. In the analysis, different text mining techniques, data cleaning and preprocessing, descriptive text statistics and sentiment analysis of announcements, and machine learning applications are used and contributions to policy making process are discussed.
How does sentiment in Monetary Policy Committee (MPC) announcement affect the exchange rate in Turkiye?
Changes in the global financial structure necessitated the shift of monetary policy instruments and increased the transparency of central banks. Thanks to advances in analysis methods, text analysis methods were improved and non-digitizable data types became digitizable. In this direction, the sentiment scores were obtained by examining the decision texts of the Monetary Policy Committee (MPC) with the sentiment analysis methods. After examining whether the mentioned text sentiment scores influence the exchange rate using the event study analysis. Under this background using whole sample estimates and rolling window estimates is to understand whether the effect is time-varying. The study used 191 Monetary Policy Committee (MPC) decision texts between the 2006-2022 period published by the Central Bank of the Republic of Turkey (CBRT). Estimation results show that the measured MPC meeting sentiment influences the USD/TRY exchange rate. Sentiment analysis indicators can be used to estimate the change in the future interest rate and support future central bank decisions and manage expectations. Furthermore, it can be analyzed whether the published MPC texts reflect the change in interest rates. The sentiment index can be used as an indicator for other econometric models that increase the explanatory power of the model.
Türkiye's export competitiveness by province: Forecasting with machine learning
The importance of export competitiveness for development has increased. The efficiency of the strategies, policies, and incentives should be carefully considered for export competitiveness. Within the scope of this study, the export competitiveness of Türkiye was analyzed by province and manufacturing sector. İstanbul, İzmir, and Kocaeli stand out regarding Türkiye's exports, which were determined as pilot provinces for the competitiveness analysis. Also, traditional and machine learning techniques for forecasting were examined and the export of provinces was forecasted by arima, boosting, and neural network. Comparing the accuracy of the models, arima was selected for competitiveness analysis. Competitiveness analyses were conducted with RCA index. The data at 2-digit levels according to ISIC Rev.4 were used from TurkStat. Accordingly, while İstanbul was competitive in 12 sectors, 2 high, 4 medium-high, 2 medium-low, and 4 low technology, in 2021, it is estimated that İstanbul will be competitive in 9 manufacturing sectors in 2024. İzmir was competitive in 2 medium-high, 3 medium-low, and 6 low technology sectors, and it is predicted that it would be competitive in 8 sectors in 2024. Kocaeli was an export competitor in 1 high, 2 medium-high, and 4 medium-low-technology manufacturing sectors in 2021 while it would be competitive in 6 sectors in 2024. It is suggested that this study, which is carried out by sector, should be expanded with analyses on the basis of product/product groups for all provinces in Türkiye. Hence, especially incentives and strategies should be evaluated in line with the findings.
Clustering analysis on G20 countries' climate change efforts regarding the energy sources
Over the last few decades climate change and its effects on the planet as well as mankind have been the topic of international debates and agreements. There has been an increasing emphasis on switching from fossil fuels to renewables as well as transforming the economy to a circular economy model. The purpose of this study is to construct a clustering analysis on G20 countries to suggest similar policies for renewable energies to the countries that fall under the same group. First, missing value imputation on the dataset was performed and then yearly data were averaged to create a single data point for each variable. Next, to account for the 80% of the variation of the dataset Principal Component Analysis was performed. Among the clustering analysis methodologies, k-means and hierarchical clustering was chosen to evaluate the groups of countries. The results of this clustering analysis showed similar groupings for the G20 countries. This study aims to show an alternative approach to policy-making procedures regarding coal phase-out and net zero goals that covers multiple countries while taking the differences between these countries into account.
Optimizing large language models: A comparative study of retrieval-augmented generation, fine-tuning, and hybrid approaches
In the rapidly advancing field of natural language processing, optimizing large language models (LLMs) for specific applications presents a crucial and intricate challenge. This paper undertakes a comprehensive comparative study of two principal optimization techniques—Retrieval-Augmented Generation (RAG) and fine-tuning— alongside a novel hybrid method that merges these approaches. The study evaluates the efficacy of these strategies using the Stanford Question Answering Dataset (SQuAD) (Rajpurkar et al., 2016). Our analysis explores the RAG method, which utilizes external data to enhance model responses, potentially easing computational demands during inference. In contrast, fine-tuning directly modifies model parameters to boost contextual accuracy and relevance, aligning the LLM's responses more closely with specific tasks. Additionally, our research examines a hybrid model that combines fine-tuning's contextual advantages with RAG's robust data-driven approach, aiming for an optimal balance between accuracy and computational efficiency. The experimental results delineate the strengths and weaknesses of each method. Fine-tuning emerges as particularly effective for tasks demanding high semantic precision, while RAG offers a scalable solution by reducing computational resource requirements. Although the hybrid model does not surpass fine-tuning in semantic accuracy, it demonstrates significant potential by providing balanced performance, which may be advantageous in scenarios requiring both efficiency and accuracy. The findings suggest tailored applications of each method depending on the specific requirements of the task and computational limitations. This study not only illuminates the nuanced trade-offs involved in selecting an LLM optimization strategy but also provides a foundation for further research and practical implementations in a variety of real-world settings.
Comparison of machine learning methods in predicting female labor force participation rates
The aim of this thesis is to determine which of the machine learning methods performs the best in predicting women's labour force participation rates and to determine which factors are the most significant in women's labour force participation. For this purpose, the second, third, fourth, fifth, sixth, and seventh waves of the World Values Survey which cover the period 1990-2022 are analyzed. Logistic Regression, Naïve Bayes, KNN (k-Nearest Neighbours), Decision Trees, SVM (Support Vector Machines), Random Forests, GBM (Gradient Boosting Machines), and XGBoost were used as machine learning algorithms. The methods used were compared based on performance metrics such as Accuracy, Precision, Recall, F1-Score, AUC, and ROC values. The results show that the best-performing methods are XGBoost and Random Forest. Along with these results, machine learning methods have shown that a woman's education level, age, number of children, and income level are the most important factors in her labour force participation decision, while a woman's patriarchal level and religiosity have limited effect on her labour force participation decision.
Comparison and evaluation of large language models (LLMS) in mathematics education
The use of large language models (LLMs), which are among the generative artificial intelligence (GenAI) systems rapidly increasing in the education sector, has become critical and important. It has become vitally significant to evaluate and compare LLMs in the context of mathematics education. This study comprehensively investigates the mathematical problem-solving capabilities of four distinct LLMs, with a focus on both final answer accuracy and step-by-step reasoning patterns. Using the Grade School Mathematics (GSM8K), a widely used dataset in mathematics education, as the benchmark, we compare the performance of multiple instruction-tuned LLMs (Gemini 2.0 Flash, Llama-3-8B-Instruct, Qwen2.5-Math, and Mistral-7B-Instruct). Our results unveil a contrast in final answer accuracy, where one model achieves near-human reliability while others cluster around substantially lower accuracy scores. In order to probe these discrepancies, we analyze step distributions of model responses and contrast them with the intrinsic structure of GSM8K. Although the dataset is naturally centered on succinct 2 to 4 step solutions, the model responses show varying tendencies: some systematically inflate reasoning chains beyond the required step count, while others collapse prematurely without providing a final numeric response. The verbose and under-reasoning failure modes are highlighted by these distributions, demonstrating that accuracy by itself is unable to fully convey the subtleties of LLMs' mathematical reasoning abilities. Together, our results highlight the potential and drawbacks of existing methods, showing that accurate mathematical reasoning necessitates alignment with the inherent cognitive problem structures.