122
Archived Theses
0
DOIs Assigned
0%
DOI Rate
Discipline
A computational perspective on meta-Fibonacci recurrence relations
In that Phd thesis, we explore meta-Fibonacci recurrence relations concept in a new approach thanks to detailed computational investigations that are provided by a computer algebra system PARI/GP. It is well known that nested recurrence relations are considered one of the most resistant subclasses of nonlinear recurrences to known proof techniques of difference equations. In that direction, we introduce and analyse not only selected spesific variants but also certain solution families while we prove their properties based on new types of structural connections between systems of nested recurrence relations. In particular, we provide remarkable computational evidences on the existence of infinitely many different solutions for a notable class of nested recurrences while this is also provable for certain families of slow solutions. Additionally, generational methods for erratic meta-Fibonacci sequences, and related statistical properties of successive generations are studied in the scope of the thesis. Keywords: Hofstadter sequence, Hofstadter-Huber family, nested recurrences
Technology and energy management for cogeneration and trigeneration systems based on multi-objective optimization in integrated structures
Since the middle of the 19th century, the world's energy consumption has increased significantly due to the rapidly growing world population and improving quality of life after the industrial revolution. Global warming is one of the most critical problems for the world's future due to the high share of fossil fuels covering the ever-increasing energy demand and the carbon emissions released into the atmosphere. Although renewable energy sources seem to be an alternative to the solution of this problem, the estimates show that the dependence on fossil resources will continue to increase in the near future. The importance of energy efficiency in generation as well as in consumption is evident since 71% of the energy source of integrated structures, which have a significant share in Turkey's total energy consumption, is based on fossil fuels. Although the integrated structures differ according to their types, they are similar in terms of energy demand consisting of electricity, heating, domestic hot water, and cooling. Cogeneration and trigeneration systems are among the most effective alternatives to meet this demand. When the studies in the literature are examined, it is seen that one-dimensional solutions such as energy efficiency or energy saving are generally developed for a particular type of building. Unlike the examples in the literature, this thesis study aims to develop a valid model for all integrated structures by considering the energy requirement of integrated structures on a common denominator, and process analyses based on thermodynamic parameters have been examined on the basis of the first law of thermodynamics. In this context, it has been aimed to define the effective working intervals of the operational optimization of cogeneration and trigeneration systems depending on the genetic algorithm. Within this frame of reference, techno-economic analyzes of cogeneration and trigeneration options for the reference integrated structures have been made using the first law of thermodynamics analyses. In the study, the most effective operational processes have been defined as one generation using the multi-objective optimization method based on a genetic algorithm. In this process, measurable technical, economic, environmental, legislative, and social criteria and boundary conditions depending on the building type have been determined. The results of the multi-objective optimization showed that the problem does not have a single analytical solution; on the contrary, it contains a Pareto set of solutions depending on the objective function, constraints, and variability. According to the analysis results obtained from case studies consisting of a hospital, hotel, and shopping mall; 26-28% from primary energy sources, 25-34% savings in energy costs, and carbon dioxide emission savings of 450-500 tons/year, 230-280 tons/year, 72-78 tons/year, respectively, have been achieved. In addition, the Sustainable Heat Management Index, which was formed as a concrete contribution to the literature, has been developed to determine the priority of the heating or cooling system in the utilization of the heat released in the trigeneration system. This thesis consists of seven chapters. In the first chapter, there are analyzes on the conceptual information of the thesis, while a comprehensive literature research and publication review based on keywords are presented in the second chapter. Energy management and technology management for integrated structures, which form the main structure of the thesis, are discussed in the third and fourth chapters, respectively. The framework model developed in the study and the related methodological infrastructure are given in the fifth chapter. In the sixth chapter of the study, the techno-economic assessment and optimization results based on the framework model formed for the integrated structures taken as reference are explained in detail. In the conclusion part, specific findings obtained from the analyzes are given, and predictions and evaluations related to them are made. At the end of the study, suggestions regarding the issues to be addressed in the future are presented.
Dynamical analysis of chaotic circuits with computational algebra method
This study analyses a novel autonomous non- linear continuous-time system derived from Bao's simple third-order inductor-free memristive chaotic circuit for chaotic dynamics via computer algebra. Bao et al. designed the Chua memristor emulator with resistors, capacitors, and opamps to avoid the negative effects of the inductor [1]. This simple third-order inductor-free chaotic circuit utilizes reasonably functional discrete components, which are significant with the number of dynamically simplified elements for mathematical representation of chaotic behaviour. Our proposed model aims to overcome the negative long-term effects on the electrical properties of the capacitor and reveal experimental results depending on an additional variable that behaves like an electrical inductor. Numerical examples are given to study bifurcation properties in the modified chaotic circuit concerning the results with Lyapunov regimes and also to identify limit cycles of period-1, period-2, period-4, and chaotic attractors in the proposed circuit.
Chaotic system analysis with genetic algorithm approach
Dynamical analysis of epidemiological models gained significant attention from applied mathematicians after the Covid-19 pandemic caused a global catastrophe. Although studies involving the decision of the suitable model to describe the epidemiologic data and the predictive analysis of the disease spread have been carried out in-depth during this phase, studies that will expose the unrecognized dynamics through predictive analysis by investigating the algebraic properties of the models such as the invariant spaces, the center manifolds, and the Lyapunov functions in the general parameter space, still have the potential to contribute vastly to comprehending the spread dynamics. This work implements the triangular probabilistic distribution to an epidemic delay differential equation as a distributed continuous delay to determine the recovery rate based on characteristic model parameters. In this way, we propose a way to reduce the model parameters that need to be obtained during model fitting by defining generic coefficients derivable from field research and population characteristics.
Comparison of statistical and machine learning methods for prediction LME metal prices
It is intended to provide a comprehensive analysis of the daily zinc prices around the world in the wake of the Covid-19 pandemic and the impact it has had on global financial zinc markets. Through the implementation of ARIMA modelling on the original zinc prices, we have gained valuable insights into the complex dynamics. According to the results, ARIMA works efficiently when attempting to fit the stationary data than the other models. This study has provided an important contribution to the existing body of knowledge regarding pandemic-related factors in financial markets assessment and strongly highlights the need to integrate such factors into the assessment of these markets as a result of the results of this study. We also apply chaotic time series analysis to the LME zinc data to demonstrate the interaction with chaos theory.
Robust high-frequency solvers
This work concerns the numerical solutions of the direct obstacle scattering problem in $\mathbb{R}^2$. To this end, we formulate the problem as an equivalent integral equation. We then review the fundamental numerical methods such as Nyström, collocation and Galerkin methods, for integral equations of the second kind. We establish convergence results and error estimates for these methods, and incorporate numerical examples considering different integral equations. Although these methods are very efficient for low frequencies, they can not be utilized for high frequency scenarios as the computational cost grows linearly with the wave number $k$. In this connection, we propose a robust convergent algorithm based on a Galerkin formulation utilizing the geometrical optics ansatz to adapt the approximation spaces to high frequency scattering (by convex obstacles) which overcomes this type of growth in complexity requiring only $\mathcal{O}(k^{\varepsilon})$, $\varepsilon>0$, increase in the degrees of freedom to maintain a given accuracy. Numerical experiments demonstrate the efficiency of our method by exhibiting numerical errors and condition numbers in several scenarios.
Numerical modeling of sloshing using the radial basis function collocation method
A series of numerical models have been developed to study sloshing in 2D and 3D Rigid Tanks. The problem is approached with Potential Flow Theory and Laplace?s equation is solved with Neumann type lateral and bottom boundary conditions and with Dynamic and Kinematic free surface boundary conditions. For the numerical method to solve the initial-boundary value problem; Meshless Radial Basis Function Collocation Method is used and a predictor-corrector type time marching scheme is used with Adams-Bashworth predictors and Adams-Moulton correctors. The 2D models were verified by performing some numerical tests and comparing the results with available second or third order asymptotic analytical formulae from the literature, numerical experiments from the literature, and with lab experiments from the literature. For the 3D models, linear analytic formulae were used for verification. The results were supported with laboratory experiments, and by employing another numerical model developed using the Virtual Mass technique. Fully nonlinear boundary conditions with deformable boundaries have been used in the 2D sloshing models. The numerical tanks have been tested with linear standing wave, lateral excitation, vertical excitation and combined lateral and vertical excitation inputs. Fast convergence and excellent accuracy is obtained for the low amplitude test cases. For the high amplitude sloshing cases, some experimental techniques are used and a good accuracy is achieved. For the 3D sloshing studies, linearized mathematical formulation was used; and the RBFCM, Virtual Mass, and the Laboratory Experiment models were benchmarked by demonstrating their performances in estimating the sloshing mode shape frequencies. Finally, a sample tank from the industry was analyzed by applying real earthquake motions using the RBFCM and Virtual Mass models, and the calculated maximum wave amplitudes were compared. The numerical experiments show that the models created in this study may be utilized for the research on the optimum tank design; which may eventually lead to improvements on the design code.
Gridification of earthquake location finding and regional fault plane solution applications
In this thesis, two studies have been conducted. One of them is Gridification of Earthquake Location Finding (ELF) Application and the other one is Gridification of Regional Fault Plane Solution (FPS) Application. Earthquake Location Finding application finds the hypocenter of an earthquake by using the seismic waveform data generated by seismic stations. ELF is data intensive since it needs to process waveform files from several stations. The application is parallelized by decomposing the problem space and expressing it as a parallel JDL workflow. To implement spatial decompositions, stations are divided among the worker nodes. A node is assigned a task which is responsible for accessing a specific station?s waveform file and computing picks. The results from all worker nodes are then sent to a collector node which runs the HYPO71 program to locate hypocenter of the earthquake. Timing results indeed show that parallelization of file accesses improves performance. The main goal of Regional Fault Plain Solution (FPS) application is the determination of the faulting mechanism that characterizes earthquake parameters (dip,strike, rake) by utilizing regional Moment Tensor Inversion method. The most time-consuming part of the FPS application involves the calculation of Green?s functions. Therefore, calculation of the Green's functions has been parallelized. A Job Description Language workflow is automatically created based on the initial parameters of individual crustal models that are necessary for producing Green functions. Timing results obtained indicate that the FPS application did achieve speed-up as a result of parallelization.
Climate change impacts on migration in the vulnerable countries
This research focuses on the economic, demographic and environmental drivers of migration related with the sustainable development in the underdeveloped and developing countries, which are the most vulnerable to the climate change impacts through the panel data analysis. We have studied some countries namely Bangladesh, Netherlands, Morocco, Malaysia, Ethiopia and Bolivia. We have analyzed these countries according to their economic, demographic and environmental indicators related with the determinant of migration, and we tried to indicate that their conditions differ according to all these factors concerning with the climate change impacts. In order to achieve sustainable development by preventing or decreasing environmental migration due to climate change impacts, these countries need to maintain economic, social, political, demographic, and in particular environmental performance. This thesis argues about climate change impacts in the future to the vulnerable countries and its effects on the migration in these countries. There are some significant risks stemming from climate change, which is not under control. When the economic and environmental conditions are considered, we have to regard that climate change will be more destructive for those who are less defendable in terms of all risks and impacts of uncontrolled climate change. In this case, there are some crucial questions required to be responded. What should be the economic and environmental priorities? What are the contributions of these priorities to the vulnerable countries concerning with the climate change effects? Considering all these, geographic structure and also human population density (how and where population live) of these poor are very significant concepts.
The projection of mean temperature and precipitation for Ethiopia, Bangladesh, and Netherlands
Climate change is an important problem for today and it will also continue to be an important problem for the future. The purpose of this thesis is to analyze the future of the climate through climate modelling in 3 countries which are geographically signifi- cant, namely Ethiopia, Bangladesh and the Netherlands. First of all, we simulated todays climate for the years between 1970 ? 2000 by using RegCM 4.3.5.5 which is the regional climate modelling for analysis of regional climate change. We used ECHAM5, which is the atmospheric global climate model, as the forcing data when making this simulation and analyzed average surface temperature and precipitation variables. We compared model simulation for 1970 ? 2000 obtained by using RegCM 4.3.5.5 with the data from CRU which is the observational station data to examine correctness of our model. As a result, it was concluded that geographical features, climate events and topography have a signifi- cant effect in climate modelling as well. Differences which occurred especially in modelling average precipitation volume indicated us that it is more difficult to model precipitation than temperature. Following the verification of todays data for Ethiopia, Bangladesh and the Netherlands, data of ECHAM5 A1B scenario was used and thus seasonal average sur- face temperature and precipitation values obtained for the climate model for the years of 2020 ? 2050 being the future time by using RegCM 4.3.5.5. As a result of comparison of todays and futures data, it was found that considerable increases will occur in temperature between the years of 2020 ? 2050.
Application of mixed coarse-grained elastic network model to protein-ligand interactions
The aim of this thesis is to explore the effect of ligand binding on vibrational dynamics of proteins specifically the collective modes that are biologically relevant.
Computational analysis of cyclonic separation
Cyclonic separation is a method to remove particles from a fluid flow using a vortex for separation. This method of separation can be used to separate fine droplets of liquid or particles from a gas flow. This device is a kind of stationary mechanical device that uses centrifugal force to separate any kind of particles from a fluid stream. There are two fundamental types of air cyclones according to the direction in which the cleaned gas leaves the cyclone. The reverse flow cyclone is one of them and it is actually the most frequently used type in industry. While the clean gas leaves from upper lid, separated particles can exit through a bottom apex. Cyclones have simple manufacturing, low energy requirements, capability to work at high pressures and temperatures. In spite of the simplicity in manufacturing and operation, the mathematical formulations used for anticipating the separation efficiency are very complicated. In contrast to most of the previous studies which use a single input, in this study two slot type cyclone separators are considered and compared together, for both laminar and turbulent flows. In this research, mathematical modeling of cyclonic separation is based on Muschelknautz Method formulation (MM) and the computational fluid dynamics (CFD) analysis is performed using the COMSOL Multi-Physics program. Some of the well-known models such as Muschelknautz D type, Swift High Efficiency, Stairmand High Efficiency and Lapple General Purpose with single and double inputs are analyzed and compared. A model with different input sizes with single and double inputs is introduced and parameterized in terms of pressure drop profile, velocity magnitude, number of particles that each cyclone can separate, separation efficiency and stream lines in laminar and turbulent streams. The particle tracing module of COMSOL is used for that purpose.
Investigation of allosteric coupling in beta-2 adrenergic receptor through molecular dynamics simulations
The purpose of this thesis is to investigate the allosteric coupling between intra- and extracellular parts of G-protein coupled human β2 adrenergic receptor (β2AR) in the presence of intracellular loop ICL3, which is missing in all crystallographic structures. In a recent study, a 1 μs MD run showed that closure of the G-protein binding site due to tight packing and closure of ICL3 under the receptor was coupled with the expansion of the ligand binding site. In this study, two independent 500 ns runs, which started from the final snapshot of the original 1 μs MD run with closed ICL3, showed that ICL3 stayed mostly in closed position, pointing out that it is a stable state blocking the binding cavity of G-protein. This allosteric coupling between ICL3 opening/closure and ligand binding site compression/expansion was further investigated by imposing specific distance constraints on the ligand binding site of the receptor. A total of seven constraints were applied between the residues on helices 3, 5, 6 and 7 that form the binding pocket. Specifically, these constraints were between residue couples Ser203Oγ-Asp113Cγ, Ser204Oγ-Asp113Cγ, Ser207Oγ-Asp113Cγ, Asn293Cβ-Asp113Cβ, Phe289Cβ-Asp113Cβ, Asn312Cβ-Asp113Cβ, and Phe289Cβ-Asp312Cβ. Constraining the binding site to an open position forced ICL3 to close from its fully open position. Constraining the binding site to a relatively closed position kept ICL3 around its open form. Using one extra constraint on backbone atoms (between residues Ser207Cα-Asp113Cα of helices 5 and 3) facilitated the opening of the binding site, and thus accelerated the closure of ICL3 from a relatively open position. However, attempting to compress the binding pocket did not lead to a fully open ICL3. Through this work, the allosteric coupling between the ligand binding regions in the extracellular parts of H5 and H6 and the intracellular part of the receptor including ICL3 was demonstrated one more time.
Meshless modeling of wave induced flows inside sandy sea beds
Ocean waves have been research focus for scientists and engineers for the last several decades, being one of the most dominant physical phenomena that affect the coastal zones. Varieties of wave theories have been developed to understand the complex nature of this phenomenon. With the emergence of computers in the midst of twentieth century and the introduction of computational approach in science, ocean wave modeling has become an integral part of the studies on ocean waves. Among several numerical methods offered, developed and applied to the ocean wave problems, one of the most recent and practical is a meshless method called Radial Basis Function Collocation Method.The motivation behind the proposed modeling effort is the development of an understanding of how oxygen entrained in coastal waters penetrates into sandy sea bottoms. To understand the related phenomena, a linear wave propagation model over a porous sandy sea bed was developed using a meshless method, the Radial Basis Function Collocation Method (RBFCM). The Green's function of the Laplace equation was used as the Radial Basis Function (RBF). The solution of equations on the boundaries was used to obtain the values inside the solution domain using the RBF interpolation. The results indicate that the method used here, a boundary-only RBFCM solution, is an efficient method for solving the problem at hand.
Numerical analysis of multiplicity and transition phenomenon in natural convection
This research is devoted to the numerical investigation of the transition and multiplicity phenomena in natural convection. The numerical simulation is performed for a 2-D closed domain using FLUENT, a commercial computational uid dynamics package. In this study, the effects of different parameters such as Rayleigh number (Ra), aspect ratio (AR), and Prandtl number (Pr) are numerically analyzed. The complex phenomenon of onset of instability is studied to find the critical Ra numbers and bifurcation points. In a specific range of critical Ra numbers, transition and multiple results are observed using Rayleigh, boundary condition and pseudo-transient continuation methods. This method uses higher order temperature boundary conditions as an initial value on vertical walls. Multiple steady state results are observed at different Ra numbers. In a range of Ra numbers, the transition and multiple steady state results can completely change the convective ow pattern within the domain.
Affect analysis of communication texts in Turkish language
As digital technologies expanded our range of online communication, textual affect analysis become the focus of considerable interest in several disciplines. While many studies have been conducted in English language, there are only a few tools specific for the Turkish language. One reason for that is the lack of lexical resources for affect analysis and annotated corpora for an accurate evaluation. In this thesis, we develop an approach for continuous and dimensional affect analysis of Turkish communications by combining several tools. We conduct experiments on various text corpora in Turkish including online multi-party chat records, psychotherapy records, Twitter data, movie reviews, and teachers' comments on high school students. Analyzing such texts brings challenges like non-standard word usage, grammatical irregularities, abbreviation usage, and spelling mistakes. We propose several pre-processing steps to deal with these. Then, we adapt an affective word dictionary from English to Turkish, and by expanding it with synsets, obtain 15,200 words with annotations for valence, arousal, and dominance. We also employ a list of frequently used abbreviations, emoticons, interjections, modifiers (intensifiers and diminishers), and other linguistic indicators to capture the overall affective state at the sentence level. We recruit and train annotators to obtain affective ground truth specifically for multi-party chat records. Our results show that the proposed system is useful, yet there is much room for improvement at different stages.
Machine learning methods in natural language processing
There is a large number of algorithms for keyword extraction and text summarization in natural language processing, as we discuss some of these in this thesis. We started with a survey on automatic text summarization in order to understand the state of the art methods. Also we proposed a new and efficient method for keyword extraction task using Word2Vec and PageRank algorithms. In this thesis, we investigated two different graph based text summarization algorithms for both single and multi-document settings on different types of texts where we used LexRank for multi-document summarization and TextRank for single document summarization. We also investigated a number of keyword extraction methods. Almost every keyword extraction method use high dimensional vectors to define words in a vector space. We approached the problem of automatic extraction of keywords from text as a unsupervised learning task and we treat each word in the document as a low dimensional vector. We developed a new keyword extraction method using Word2Vec and PageRank algorithms. Our results show that summarization algorithms give best result on news texts, usable results on legal texts while they give less than optimal results for short stories. On the other hand, we also compared differences in using one-hot-representation and Word2Vec representation but we observed no significant differences between these methods.
Facial expression recognition in the wild using improved trajectories and fisher vector encoding
Automatic video data analysis has been a growing interest in order to improve human computer interaction. One of the most challenging parts in video analysis is the ability of evaluating human emotion robustly. Vast applications of human facial expression recognition can be seen everywhere from educational systems to treatment of Asperger's and surveillance. In this thesis, we explore facial expression recognition on both laboratory and realistic videos. After studying recent works about face detection, facial alignment, video description and classification, we present our novel approach in, which our proposed pipeline including facial alignment in combination with improved dense trajectory, geometric, encoded with Fisher vector encoding and LGBP-TOP features are fed to extreme learning machine. It is the first time that improved dense trajectory features are used in facial expression recognition. Furthermore, we extensively study each step of our pipeline in a comparative manner. We evaluate our approach on CK+ and EmotiW 2015 challenge datasets. Videos in first dataset are captured in laboratory settings and start from neutral state and end with peak expression while the second one is selected from movies with realistic conditions, spontaneous emotions, complicated background and challenging illumination variations. On Ck+ dataset, we obtained 94.80\% and 95.79\% (without contempt) accuracy, which is among the best results obtained on the CK+. On EmotiW 2015 challenge dataset, we got 43.39\% accuracy, which is higher than the baseline of the challenge considerably. In both datasets we were able to obtain the state-of-the-art results. Our results show that using appropriate pipeline of face alignment combined with efficient visual descriptors can result in a robust system with high ability of recognition.
Determination of local wind energy potentials using climate modeling
After the industrial revolution, the needs of coal and other fossil fuels has been increasing day by day. Intensive burning of fossil fuels has resulted in a rise in the atmospheric CO2 concentration. In the 18th century, the amount of CO2 in the atmosphere was almost 280 ppm and now this value is more than 400 ppm. As a result of that, climate is changing rapidly and predicting the climate and its variables, such as wind speed and direction, is more difficult than before. It is more important to use renewable energy to reduce CO2 emission to the atmosphere. To be able to do, wind energy is a good option and it is a rising star in the renewable energy market. Determining location of wind power plant according to wind direction and wind speed of region is vital for market. In order to solve these kinds of problems, an alternative method is applied in this study. To predict wind direction and speed according to the climate change, firstly, I used RegCM4, one of Regional Climate Model version of The Abdus Salam International Centre for Theoretical Physics (ICTP), which is developed in ICTP, for simulating the climate variables and, then I used an open source software Winds on Critical Streamline Surfaces (WOCSS), which is developed by Dr. Francis L. Ludwig, for projecting the wind direction and speed. The result of this study shows wind speed will increase at the Canakkale Biga region at the future.
Sfingozin kinaz 1 (SK1)'in ortosterik ve allosterik inhibitörlerinin in siliko yöntemlerle belirlenmesi
The sphingosine kinase 1 (SK1)/sphingosine-1-phosphate (S1P) signaling pathway is a crucial target for numerous human diseases from cancer to cardiovascular diseases. Even though the discovery of SK1 inhibitors has been considered as effective chemotherapeutic agents, early SK1 inhibitors that target the ATP or lipid-binding sites have suffered from poor potency, selectivity and pharmacokinetic properties. Thus, there is an urgent need for the discovery of in vivo stable orthosteric SK1 inhibitors with improved potency and selectivity. Also, the selectivity issue of the kinases, which share highly-conserved ATP-pocket, can be overcome by targeting the less-conserved allosteric sites. SK1 is known to function minimally as a dimer; however, the crystal structure of the SK1 dimer has not been determined. In this study, a template-based algorithm implemented in PRISM was used to predict the SK1 dimer structure and then the possible allosteric sites at the dimer interface were determined via SiteMap. Both the predicted putative allosteric site(s) and the lipid-binding site were used in an integrated workflow of ligand-based and structure-based pharmacophore modeling, virtual screening, molecular docking, re-screening of common scaffolds to propose a series of compounds with different scaffolds as potential orthosteric and allosteric SK1 inhibitors. The stability of the SK1-ligand complexes was analyzed by molecular dynamics simulations. As a final outcome, ligand 5 having a 1H-benzimidazol-2-ol scaffold was found to be a potential selective orthosteric inhibitor as well as ligand 7 having a 4,9-dihydro-1H-purine scaffold and ligand 12 sharing a 2,3,4,9-tetrahydro-1H-β-carboline scaffold were found to be potential selective allosteric inhibitors for SK1.
Metabolic modeling of daphnia for environmental toxicity studies
Daphnia, a genus of small planktonic freshwater crustaceans commonly referred to as "water fleas," plays a critical role in aquatic ecosystems and has long served as a model organism in ecotoxicological studies. Owing to its sensitivity to pollutants, short life cycle, ease of laboratory cultivation, and transparent body, Daphnia pulex has become a widely adopted test species for acute and chronic toxicity assessments. As a heterotrophic organism, Daphnia's growth and survival are highly dependent on food availability and quality, particularly essential nutrients such as phosphorus, polyunsaturated fatty acids (PUFAs), and sterols. Among various feed sources, Chlorella vulgaris has demonstrated considerable nutritional value for Daphnia cultivation, exhibiting high protein content and the ability to effectively remove selenium from wastewater. In this study, a genome-scale metabolic model of Daphnia pulex was constructed using the COBRA Toolbox within MATLAB, enabling simulation of its metabolic behavior under different nutrient conditions. The model was employed to evaluate the organism's growth response to various dietary inputs and to identify key metabolic pathways involved in nutrient assimilation. In the final stage, toxicological assessments were carried out by integrating metal concentration data from Lake Beyşehir with the metabolic model. Heavy metal compounds commonly found in freshwater environments were analyzed in terms of their potential impact on Daphnia's metabolic performance. The study presents a comprehensive framework that combines metabolic modeling with toxicological evaluation, offering novel insights into the metabolic adaptability of Daphnia pulex under both nutrient-driven and chemically stressed conditions. The results support the potential use of this integrated modeling approach for predicting the ecological consequences of pollutant exposure and assessing algal-based feed viability in contaminated water systems.
Rarest precipitation events in Turkiye willsee the greatest relative increase inmagnitude under future climate change
Changes in precipitation regimes occur due to global climate change, and these changes may occur in the form of increases in extreme precipitation events and drought as a result of decreased precipitation. This study undertakes a critical examination of T¨urkiye's rarest precipitation events and their anticipated changes under future climate change scenarios, employing the Regional Climate Model (RegCM4.3.5) and focusing on RCP 4.5 and RCP 8.5 scenarios. It evaluates precipitation in T¨urkiye for return periods of 1, 10, 100, 1000, and 10000 years during the periods 2026-2050, 2051-2075, and 2076-2100, based on MPI-ESM-MR scenario predictions under CMIP5, tailored for the region. Employing a range of distributions, including log-normal, GEV, and Gumbel, the log-normal distribution was found most accurate, particularly for the 2026-2050 RCP8.5 scenario. Under the RCP4.5 scenario, the Mediterranean region is expected to see a significant increase in extreme precipitation events with 100-year return periods, while Central Anatolia may experience a decrease. Under the RCP8.5 scenario, a substantial intensification of extreme precipitation events is projected for the Mediterranean and Aegean regions throughout the century. These findings highlight the urgent need for region-specific adaptation strategies in water management, flood defense, and sustainable urban planning, especially in the most affected Mediterranean and Aegean regions. This research not only offers important insights for future policy and research but also emphasizes the need for proactive measures in response to evolving climate patterns.
Identification of biomarkers for cancer by machine learning
Cancer is one of the deadliest diseases and early treatment is very important for successful recovery. However, early detection is required for early treatment, and currently, it is not an easy task for a lot of developing countries. Using biomarkers to fasten the diagnosis process may help to the solution of this since they can be detected through bodily fluids easily. Also, finding genetic factors leading to cancer is another task that is difficult to accomplish due to cancer being a complex disease as well as genetic data being a big data. Biomarkers detected from the healthy and cancerous tissue datasets may be informative for further research on this topic. In this thesis work, we developed an algorithm to detect biomarkers for cancer. Several machine learning models are used with the help of the PCA dimensionality reduction technique in the algorithm. The proposed algorithm is also converted to a publicly available software PREDIGENE which can be used by uploading a CSV file containing gene expression counts of healthy and diseased tissues. PREDIGENE is tested on two different cancer types (breast and liver) with six different datasets in total, three for each cancer type. The relevance and consistency of the results PREDIGENE returned are examined through a literature search and comparison with another publicly available tool developed for the same purpose. In conclusion, PREDIGENE, our tool to find cancer biomarkers through the usage of artificial intelligence, shows promising results when compared to the current literature.
The performance of different parametrization on RegCM
This study investigates the accuracy and reliability of climate predictions utilizing the RegCM 4.7.0 regional climate model, focusing on two distinct topographies in Türkiye. The primary objective is to identify which physical parameters enhance the model's ability to simulate climate scenarios and to assess how varying resolutions impact computational efficiency without compromising accuracy. This study employs precipitation and temperature data for dynamic downscaling, analyzing high and low topographic areas at 25 km, 10 km, and 5 km resolutions. Appropriate parameters, such as Planetary Boundary Layer and Convective Precipitation Schemes, are selected. The results are then compared with CRU and NCAR reanalysis data, which are interpolated to match the model's resolution. The findings reveal that temperature bias values are more accurate than precipitation values, and the same physical parameters yield better results in high topographic areas. An increase in resolution corresponds to an increase in bias values. Spatial correlation values for temperature decrease as resolution increases, showing better results in higher topography. For precipitation, the best outcomes are observed at an altitude of 10 km, with no specific pattern for low topography areas. In conclusion, the study demonstrates that high topographic areas exhibit moderate correlations, while low topographic areas show little to no correlation. The research suggests that a 25 km resolution is the most efficient for RegCM 4.7.0, as higher resolutions may lead to minor deviations due to interpolation.
Scalability of projected patterns of extreme weather events over Europe using regional climate model projections
Climate change leads to significant increases in the frequency and severity of extreme weather events. Combating climate change and developing adaptation strategies is of great importance for a sustainable future. The effects of climate change need to be understood and addressed, and climate models are used for this end. However, due to differences in climate models and systematic errors in model calculations, different techniques such as "pattern scaling" are also used. The study used a pattern scaling approach, extracted from different global climate models, to examine how changes in temperature extremes and total precipitation in the EURO-CORDEX region respond to increases in global temperatures using certain extreme weather indices. According to the results, the yearly minimum value of daily minimum temperature (TNn) increases more than the yearly maximum value of daily maximum temperature (TXx) for Europe. The multi-model mean of the changes in scaled patterns of extreme temperatures emerges early, around 2020, even before it becomes robust. Additionally, the total annual precipitation amount (PRCPTOT) in the northern regions increases by up to 6-9% compared to the reference period, while decreases of up to 12-15% can be observed in North Africa. Finally, using the ERA5 dataset, the change in patterns of four extreme temperature indices was examined to determine whether end-of-century scaled patterns could be detected for the conditions that were observed. The observed patterns of change closely resemble the end-of-century scaled patterns identified by the GCM-RCM sample, indicating that end-century patterns are already emerging in some parts of the domain.
Market capitalization-based portfolio weighting in BorsaIstanbul: A comparative analysis of historical performanceagainst established benchmarks
This thesis examines the medium- to long-term performance of market-capitalization-weighted portfolios covering the entire Borsa Istanbul equity universe, adjusted for free-float shares and rebalanced at varying frequencies (1-day to 8-week). Using a comprehensive dataset spanning January 2011 to October 2024 with daily prices and actual free-float ratios, the study evaluates whether such portfolios can outperform benchmarks like BIST~30 and BIST~100. A total of 754 simulations per configuration enables robust comparisons across horizons from 12 to 172 months. Results indicate that full-market portfolios, when dynamically rebalanced, tend to outperform benchmarks in median and upper-quartile returns from the 24-month horizon onward. However, this advantage is limited in scale and does not yield significantly higher cumulative returns. Despite a nearly fifty-fold increase in total market capitalization, the portfolios captured only a modest portion and consistently underperformed the gold benchmark (XAUTRY) over long horizons. Risk metrics such as Sharpe Ratios and Maximum Drawdowns support these findings, with 10-day to 2-week rebalancing frequencies showing better risk-adjusted performance. The methodology avoids biases like overfitting, calendar effects, and survivorship. Overall, broader, systematically rebalanced portfolios appear to offer viable long-term alternatives to conventional Turkish equity benchmarks.
Anomaly detection in time series
Anomaly detection (AD) is the discovery of the observations which does not conform with the rest of the observations. The types of anomalies and their occurrences that exist in the data set are tried to be determined. On the other hand, time series structures have dynamic structures, which are evolving over time, and in such structures, observations will be affected by previous observations. This thesis focuses on the anomaly detection process under time series structures. This problem is not always straightforward because the definition of anomaly could change with the context of the dynamic structure and anomaly detection process in the system could interfere with the intense noises at the observations. In this thesis, we try to identify anomalies in the sub-sequences of the streaming data. When doing so, we also want to discriminate the anomalies in the system with the faulty observations. Therefore we investigate collective anomalies in the data. We propose both statistical inference methods and deep learning approaches for such type of anomaly detection in time series (ADTS) problem. We use a Gaussian mixture model (GMM) and a customized hidden Markov model (HMM) as statistical approaches, while we use Recurrent Neural Networks (RNN) and Long Short-Term Memories (LSTM) as deep learning approaches. Except for GMMs, we take into account the sequential structures of data sets in the models proposed above. We apply our methodologies to the Borusan wind turbines data and we compare the model results with the experiments we performed on this dataset.
The effect of wind pattern change on the environmental/climatic conditions within cities
The labor force demand by the Industrial Revolution has led to an increase in the population of cities. This situation was followed by rapid development of urban areas with starting of nineteenth century to accommodate labor force around the industry. This situation has ended up with altering more land surfaces to buildings, car parks, roads or other structure types. However, it cannot be said that all the urbanization steps are taken properly. Thus, these anthropogenic developments have affected wind flow in urban. Arbitrary built urban areas restrict the wind flow in street canyons and make it slower than the flow above the building blocks, hence reduce the cooling effect of wind at near-ground levels. However, increasing the wind speed in streets can improve thermal comfort of people by the means of convective heat transfer through the skin. Thermal comfort and wind flow patterns are important environmental issues when designing new urban areas. Starting from this point, this thesis focuses on computational fluid dynamics (CFD) models to observe wind flow and thermal comfort of arbitrary built urban areas in Mecidiyeköy, Istanbul. Mecidiyeköy is an arbitrarily urbanized and one of the most crowded hubs of Istanbul, which is modelled as it is and compared with alternative design scenarios in wind flow and thermal comfort results. In general, this thesis analyzes the impact of buildings on wind flow, hence thermal comfort in cities.
Genome-scale metabolic reconstruction of freshwater organism daphnia pulex
The aim of this study is the reconstruction of the genome-scale model of an ecological organism and investigation of its cellular functionality towards toxicity by using computational systems biology approaches. Daphnia pulex is one of the keystone species of the toxicity tests for the environment, and hence it is here chosen as a model organism. The study on D. pulex genome was first published in February 2011 and contributed to numerous scientific researches. In the present study, manually curated genome-scale metabolic model of D. pulex is reconstructed by using KEGG and UniProt databases along with the literature search. The completed model includes 1051 reactions (985 internal, 66 exchange) and 774 metabolites (222 mitochondrial, 485 in the cytosol, 67 externals) related to 697 genes. Based on the constraint-based modeling, biomass is chosen as the objective function to inspect the cellular fluxes through the growth of D. pulex. The robustness analysis is performed on selected compounds such as L-isoleucine, selenite and sulfate so as to monitor their effects on biomass. L-isoleucine has a positive impact on the growth of D. pulex. The applicability of the reconstructed model of D. pulex to toxicity studies is tested by investigating the relation between selenate and sulfate uptakes, where these fluxes are estimated by flux balance analysis. The computational results demonstrate that when sulfate uptake is getting close to zero flux, selenate uptake begins. This observation on selenate and sulfate uptake mechanism indicates that predicted results are in agreement with the literature. Consequently, the reconstructed metabolic model of D. pulex is able to predict the cellular behavior and can be used for metabolic toxicity studies.
Climate change projections of Cyprus using regcm4.4
The scientific motivation of this thesis is a substantial presence of climate change which is one of the most significant issues in the world. There have been happening many natural disasters because of climate change. In particularly, coastal regions and islands, such as Cyprus, are more vulnerable to any possible hazards. Humankind start to contribute the climate change, especially, after indus- trial revolution. Burning fossil fuels causes the increase of CO2 emission to the atmosphere. In fact, at the beginning of the 18th century, before the Industrial Revolution, CO2 concentration in the atmosphere was 280 ppm; however, today, this value is 410 ppm. The unprecedented CO2 increase in the atmosphere since Industrial Revolution gives rise to global climate change. The Mediterranean region is predicted to be affected by climate changes in terms of air temperature rise. In this study, air temperature (◦C) and precipitation (mm/day) estimations of Cyprus were investigated with the regional climate model. The climatology model is run for the three periods of 2011–2040, 2041–2070, and 2071–2100, with respect to the control period of 1971–2005 for Cyprus domain via regional climate model simulations. Regional Climate Model (RegCM4.4) of ICTP (International Centre for Theoretical Physics) was run by using two different global climate models. MPI- ESM-MR global climate model of the Max Planck Institute for Meteorology and HadGEM2 of the Met Office Hadley Center were dynamically downscaled to 10 km resolution by using double nesting. RCP4.5 and RCP8.5, the emission scenarios, of the IPCC (Intergovernmental Panel of Climate Change) were used for projections.
Statistical return periods of extreme weather events
The Middle East & North Africa (MENA) region is one of the most populated areas on Earth in terms of city population. There are 643 cities in MENA region with population of urban agglomerations with 300,000 or more in 2018 according to United Nations. In this thesis, return periods of extreme temperature events are calculated using the probability density functions. For this purpose, Global Climate Model (GCM) outputs of Max Plank Institute Earth System Model Mixed Resolution (MPI-ESM-MR) and Hadley Global Environment Model 2 - Earth System (HadGEM2-ES) are dynamically downscaled to 50 km for the MENA region by using the Regional Climate Model v4.4 (RegCM4.4) for 2 different Representative Concentration Pathways scenarios, namely RCP 4.5 and RCP 8.5. Elevation correction is applied to each point for sea level. Temperatures at city centers are calculated from the nearest 4 grid points using inverse distance squared interpolation method. Daily maximum temperatures histograms are plotted for each city and future predictions of return periods are compared with the reference period of 1971-2000 using the means and standard deviations obtained from Gaussian Mixture Model. The results show that the frequency of extreme events increases for all cities between 2070 and 2099. Peaks in temperature distribution are diverging from each other which will cause more severe extreme events. This divergence would cause cities to have shorter transition seasons and their climate would transform into only 2 seasons.
Design of donor-acceptor conjugated polymers for high-performance organic solar cells: a quantum mechanical approach
In this study, the design of donor-acceptor conjugated polymers for high-performance organic solar cells is intended by investigating electronic, geometrical and optical properties of a specific set of donors that are used in polymer solar cells. 4 different donors including benzo [1,2-b:4,5b'] dithiophene (BDT), 3,3'-difluoro-2,2'-bithiophene (diF2T), naphtho [1,2-b:5,6-b'] dithiophene (NDT), thieno[3,2-b] thiophene (TT) and 1 acceptor named 2-propyl-5,6-difluorobenzo[d][1,2,3]triazole (ffTAZ) are studied in order to set a benchmark. The 4 different donors studied for benchmarking and 7 different acceptors including benzo [c][1,2,5] thiadiazole (cop1), [1,2,5]thiadiazolo [3,4-g] quinoxaline (cop2), [1,2,5] thiadiazolo [3,4-d] pyri-dazine (cop3), [1,2,5]thiadiazolo [3,4-c] pyridine (cop4), [1,2,5] oxadiazolo [3,4-d] pyri-dazine (cop5), fluorinated naphtho [1,2-c:5,6c'] bis [1,2,5] thiadiazole (FNTz), fluori-nated [2,1,3] benzobisthiadiazole (FBTz) are combined and investigated for 28 different predictions. Geometrical and optical characteristics were calculated using Density Functional Theory (DFT). Distortion energies and reorganizational energies for electron and hole transfer were calculated using B3LYP/6-311G*. In conclusion, 2 out of 28 molecules were chosen as the most effective molecules to be used in high-performing organic solar cells. The molecules chosen contain NDT and TT as donors and FBTz as acceptor.
Gridded distribution of daily non-uniform meteorological station data of Turkey
The climate modeling is a complex area to work on. It is not easy to represent the atmospheric events in a model, because there are a lot of drivers and variables which affects the local and global climate. Even if, somehow, some meteorological phenomenon can be fully represented mathematically, it is even harder to do the calculation through a computer or a super computer with the current technological developments. So, some assumptions, optimizations and simplifications must be done to work on climate. This study will examine the climate measurement station data distributed in Turkey. The randomly distributed real station data all around Turkey gathered from the Turkish State Meteorological Service database. Our goal is to obtain uniformly distributed grid data for temperature and precipitation. We have used interpolation, one of the most ancient and common computational method with a scale that allows us to create a high-resolution temperature and precipitation map of Turkey. With the help of MATLAB build-in tools and functions, the scattered station data turned into a uniformly gridded data. The station data was very hard to work on, since the data begins from 1960. Meanwhile, the meteorology stations moved, closed, opened which caused a lot of incomplete and disconnected data. So, some normalization must be done, and the data become suitable for interpolation. To do the interpolation calculation, we used the MATLAB build-in function, scatteredInterpolant. It allowed us to calculate the scattered data into a gridded data. Once the interpolation completed, another MATLAB toolbox used to project the results on a regional map. The mapping toolbox is very powerful in printing the results on a map, in this case on Turkey regional map. Finally, we have managed to obtain Turkey temperature and precipitation station data map between 1960-2016 in daily basis.
Computational study on allostery in bacterial ribosome
Molecular machines in a cell have signal processing to perform their function using specific sites such as active or allosteric sites. Ligand binding to active sites or signal transferring from allosteric sites affect their function and dynamics. In this thesis, firstly crystal structure of bacterial ribosome (4kdk-4kdj) and its conformers which are generated by ClustENM are investigated to determine allosteric communication pathways. Targets on ribosome are determined as the Decoding Center (DC) – the Sarcin Ricin Loop (SRL), DC - the Peptidyl Transferase Center (PTC) and the PTC – Tunnel. On the allosteric pathways between DC and SRL, EF-G stands out with critical sites on its domains IV which has a significant function in blocking back translocation of tRNA. Some significant nucleotide and aminoacid like A1493 and Met580 appear on pathways between DC-SRL, which help EF-G hydrolysis. On the DC-PTC pathways drug binding site is observed. On the PTC-ribosomal tunnel pathway has a highly conserved non-Watson-Crick base pair and binding pocket for antibiotic is found. Secondly, the bacterial ribosome of E.coli, 4v5h, is analyzed to investigate allostery between Secretion Monitor (SecM)–PTC and TF-Ribosomal Tunnel. A76 of tRNA and nascent chain which consists of alanines seems significant between U2585 and A2451 to provide allosteric communication between SecM and PTC. On the shortest pathways on the TF-Ribosomal Tunnel, GLY91 from L22 has a high frequency of occurrence on all pathways. GLY91 is significant for elongation arrest and for the turn of the β-hairpin of L22 which is important since antibiotic resistance appears when a mutation on the β-hairpin occurs.
Climate model performance assessment for MENA domain
This study aims to evaluate the accuracy of global climate models developed under CMIP5 in temperature simulations, specifically for the MENA (Middle East and North Africa) region. Using ERA5 reanalysis data as a reference, the performance of a total of 32 models for the period 2006–2024 was analyzed using various statistical metrics, including correlation coefficient, normalized RMSE, normalized standard deviation, bias, and standard deviation ratio (SDE). All models were first downscaled to resolution, and then interpolated to the ERA5 0.25o grid for comparisons. Three subregions-T¨urkiye, Iran, and Saudi Arabia-were identified, and detailed comparisons were conducted at both the broader MENA scale and the regional level. Visual analyses were supported by bias maps and Taylor diagrams. The thesis also utilized the Ranking Metric (RM Score), which combines multiple metrics to provide a single evaluation metric. Each model was ranked based on its performance based on this combined score. This method prioritizes models with high correlation, low error measures, and normalized standard deviations. The results show that the CESM1-BGC, CCSM4, and CESM1-CAM5 models stand out as the best-performing models, achieving the highest RM scores in many regions, while models such as inmcm4, IPSL-CM5B-LR, BNU-ESM, and GISS-E2-R have the weakest performance with lower scores. Regional differences are also striking: the results suggest that while CMIP5 models are generally successful in capturing large-scale climate trends, model selection based on regional performance is critical for local-level applications.
Likelihood free particle filtering with approximate Bayesian computation for parameter estimation in cosmic ray air shower studies
Highly energetic particles called cosmic rays arrive at earth and interact with the atmosphere to cause other particles to emerge and fuel a chain interaction of particle production until the energy is dissipated. These cascades emerging from a highly energetic initial particle and producing secondary particles that spread over a large area at the ground are called Extensive Air Showers. These showers are studied to find anisotropy in their arrival direction and energy to unveil their source and production mechanisms in the universe. In this study we utilize the likelihood free particle filtering with Approximate Bayesian Computation to estimate the incident angle and energy of the primary particle. ABC method makes use of comparison between simulated and observed summary statistics to overcome the problem of computationally intractable likelihood function of particle physics.
Structural behavior of non-pneumatic tires using finite element analysis
Modern tire technology has revolutionized vehicle performance in terms of safety and comfort by combining improvements in production techniques, materials, and design. However, conventional tires have major drawbacks like punctures, pressure loss, and temperature fluctuations, leading to a reduction in efficiency and an increase in maintenance requirements. In order to overcome these limitations, the development of non-pneumatic tire (NPT) technology has gained importance. NPTs have been characterized by their airless design using durable materials such as solid rubber or foam with major advantages like puncture resistance, minimal maintenance, and excellent load-bearing capability. This study focuses on the structural behavior of the Michelin Unique Puncture-Proof Tire System (UPTIS) based NPT using the finite element analysis. For static vertical loading, the effects of collapsible spoke thickness, spoke angle, and reinforcement thickness on the vertical displacement, vertical stiffness, contact pressure, and rolling resistance are evaluated. Additionally, the steady-state rolling analysis of NPT is performed in order to investigate the dynamic characteristics which are contact pressure and shear stress distribution on the contact area in braking, freerolling, and traction states. As UPTIS and NPTs have been gaining more popularity in the automotive industry, the findings of this study contribute to understanding of the relationship between design parameters and NPT performance.
Modeling volatility dynamics in financial time series: Ananalysis of econometric models and machine learningmethods
In this thesis we present an analysis of different volatility dynamics in financial and cryptocurrency markets. We focus on the Borsa Istanbul stock exchange index BIST 100 (XU100), NASDAQ index, and Bitcoin/USD exchange rate (BTCUSD). Our aim is to understand the risk profiles of the data, forecast future risk using historical data, and statistically analyze structural characteristics of the volatility in the markets. In our analysis we used logarithmic returns of each financial asset we considered. Our first step of the investigation was to determine the best fitting ARIMA models, and analyze the parameters of these models using tools such as ADF-test, ACF- and PACF-plots, QQ-plots, and Kolmogorov-Smirnov test. Our working hypothesis is that if a model successfully explains the behaviour of an asset then the residual signal must be close to white noise. However, we found significant autocorrelations in the residuals which indicates that the ARIMA models we used did not fully capture all underlying patterns. In the next step, we used GARCH models to analyze the dynamics of the volatility inherent in each asset. While the risk profiles of the datasets varied, the structure coefficients of the models were consistent across all datasets. We finish the thesis by outlining how one might expand our study using endogeneous and exegeneous data, and different machine learning methods.
Evaluation of different downscaling approaches for very-high-resolution climate data
Climate change leads to widespread changes in atmospheric and oceanic conditions, increasing the frequency of climate anomalies and negatively impacting ecosystems and human communities. It is crucial to understand climate change and make accurate predictions about it. Climate change studies focus on tools like General Circulation Models (GCMs); however, GCMs cannot accurately represent local climates, leading to uncertainties due to their coarse resolution. Statistical and dynamical downscaling techniques improve local climate projection accuracy. This study compared statistical and dynamical downscaling techniques for evaluating Turkey's climate change projections, using the MPI-ESM-MR as the main GCM, RegCM4.7.0 regional climate model for dynamical downscaling and the spatial delta method for statistical downscaling. 17 datasets were analyzed to investigate spatio-temporal correlations at resolutions of 1km, 5km, 10km, and 20km. Evaluated spatial correlation of precipitation and temperature showed low to moderate correlation coefficients with negative correlations and near-zero values for precipitation but higher correlation results for temperature. 10 and 20km resolution downscaling data showed more favorable results. The temporal correlation of precipitation showed superior consistency with reduced standard deviations and improved correlation coefficients. The study highlighted the temporal correlation of temperature, exhibiting exceptional precision due to its nature and alignment with annual seasonal cycles. This study's findings will significantly enhance understanding of the optimal methodology for downscaling climate change projections and the impacts of climate change on local communities.
Üretken modellerde boyutsallik ve mimari uzerine analiz
Deep generative models are powerful class of machine learning models. How- ever, a significant amount of computing power and technical knowledge is required to conduct the training process. Even searching for hyperparameters requires a high com- putational cost. Moreover, there is still ongoing research on methods for evaluating generative models, and owing to the lack of a robust and consistent metric, there are limited comparisons between generative model architectures and algorithms. In this study, we attempted to compare two types of generative model architectures, Gener- ative Adversarial Networks (GANs) and Real-valued Non-Volume-Preserving (NVP) flows, with synthetic datasets as well as with a well known image dataset MNIST. We evaluate their data capturing ability according to data dimensionality and variability. We propose an Minimum Description Length (MDL) based metric to examine the ef- fect of model complexity which is measured as model's parameter count. We provide estimated Kullback-Leibler (KL) divergence and propsed MDL-based metric results. Our findings indicate that NVP models have the capability to encode more data variability while utilizing fewer parameters when contrasted with GANs for lower di- mensional datasets. The proposed MDL-based metric, facilitates selecting suitable architecture in terms of model complexity for a given dataset considering its variability and dimensionality.
Variance reduction in Markov chain Monte Carlo algorithms
In a simulation study, the output of the simulation is used to estimate target parameters. The estimated parameters typically has a variance, which impacts the precision of the estimation. In Monte Carlo simulations, the standard approach to increase the precision of the estimation is to increase the sample size. However, we can increase simulation size until a certain point due to resource restrictions. This is where Variance Reduction (VR) techniques help. These techniques use simulations as an input and return another simulation with almost the same expected value and less variance value. VR techniques are widely studied in the context of Monte Carlo simulations. However, the research on the application of these algorithms in Markov Chain Monte Carlo (MCMC) simulations is limited. In this thesis, we study the effectiveness of VR techniques for MCMC algorithms. We created an experiment environment that allows us to control the difficulty of the problems that are solved by MCMC algorithms. In our study, we use the Random-Walk Metropolis-Hastings algorithm with different types of proposal distributions. We first evaluate the convergence of the MCMC algorithms, select chains with good convergence properties, and then apply the VR techniques. The VR techniques of interest are {\it{general}} VR techniques, which are Antithetic Variates, Control Variates, and Latin Hypercube Sampling. We study the effectiveness of these algorithms using problems with varying difficulties. We address the problem of calculating the standard error of the MCMC simulation outputs, evaluate the accuracy of the estimated values, and we evaluate the correctness of the standard error calculations by calculating the coverage probabilities of the confidence intervals and RMSE/MAE values of the estimations. We show that VR techniques are successful at decreasing the variance of the MCMC simulations without introducing any bias.
Example-dependent cost-sensitive gradient boosting machines for credit scoring
Although most of machine learning algorithms try to minimize cost-insensitive losses, many real world applications require cost-sensitive approaches where misclassification costs among classes differ from each other. In addition to misclassification costs, examples in data sets may have nonidentical costs which is a case of example-dependent cost-sensitive learning. For example in credit scoring, mistakenly rejecting a good borrower and approving a bad client with financial distress result in different costs. Additionally, providing variety of credit amounts to applicants makes the credit scoring example-dependent. In other words, falsely approving 100M$ and 1M$ loans produce unequal costs. To overcome this problem, this thesis proposes an example-dependent cost-sensitive loss function. With the introduced loss function, cost sensitivity is handled during the learning process. This is achieved by changing the traditional loss function of Gradient Boosting Machines with the proposed one to make it Example Dependent Cost-Sensitive Gradient Boosting Machines. The proposed algorithm is tested on two real world data sets that include credit amounts and synthetically generated data sets. The algorithm is compared with cost-insensitive learners, previously proposed example-dependent cost-sensitive classifiers that handles cost-sensitivity during learning, a post-processing method called Thresholding and a pre-processing method Oversampling to make cost-insensitive classifiers cost-sensitive. Results show that our method outperforms those four methods in terms of financial savings.
Short-term forecast methodologies and case studies in traffic flow
This thesis gives case studies on short-term traffic flow forecasting strategies within a time series framework. After discussing the traditional, machine learning and deep learning methods, one of main goals is to experiment on the uses of hybrid methods. Besides analyzing approaches that were already used in the traffic flow literature, we also introduce and test distinct strategies. Further, we supplement our point forecast results with interval forecasts. In particular, quantiles regression based intervals such as quantile regression averaging and quantile regression neural network are implemented. Both point and interval forecasts are evaluated via several evaluation metrics, and an extensive comparison is provided among the methodologies studied.
Probabilistic discriminative region descriptor for time series classification
Detecting discriminivative regions is a recent promising concept in many different domains for various dataset types such as image, text and time series. In time series domain, time series might be large and high dimensional because of the developing storage capacities. Although computational capacities are improved, storage and computation costs are increased. Therefore recent attempts are focused on the decreasing the computational and run time complexities. To decrease the complexity of the models, instead of using raw data, construction of the new feature representation by using the distinctive sub-sequences of the time series is the most common approach. Discriminative sub-sequences are called as shapelets in time series reflect the characteristics of the class of time series. Shapelets provide interpretable results and shapelet-based classifiers have superior accuracy on many time series datasets. Many researchers have proposed shapelet extraction methodologies for classification purpose. This study proposes a novel local feature extraction framework for time series and shapelet-based time series classification pipeline. Proposed framework provides model selection flexibility to describe the time-observation space to find local discriminative regions. After obtaining the discriminative regions, shapelets are extracted on the time-observation space by thresholding the class probability estimates to construct a new feature representation. New feature representation is calculated by the Euclidean distance between shapelets and time series. Finally, a classifier is trained by the new feature representation. Experimental results show that shapelet-based time series classification by using proposed Probabilistic Discriminative Region Descriptor (PDRD) provides competitive results on benchmark datasets.
A tree based categorical variable encoding strategy in supervised learning tasks
Categorical variables are present in most real-world datasets, often consisting of a high number of levels, referred to as high-cardinality categorical variables. Most machine learning algorithms do not have an innate mechanism to deal with categorical variables, hence, their encoding is necessary. Categorical variable encoding is the general term for the conversion of nominal independent variables to a numerical format. Many encoding strategies exist, and they are discussed in this thesis. This thesis presents a novel encoding strategy, categorical split encoding, and also provides an analysis of existing encoding methods. Categorical split encoding uses primary and surrogate split information as the vector representation for categorical variables, through a tree-based algorithm, this method outputs binary columns for each categorical variable making use of target information. Missing values are imputed by using surrogate information, while clustering similar values together based on the path they take through the decision tree algorithm. Various existing encoding strategies are benchmarked for comparison with the proposed strategy. The performance of categorical split encoding and other encoding methods is compared with three different machine learning algorithms (generalized linear models, random forest and xgboost) using datasets from regression, binary and multiclass classification settings. Datasets used are made publicly available for replication purposes. As a result, categorical split encoding provides competitive results compared to existing encoding strategies in various datasets.
A comparative evaluation of machine learning algorithms for statistical downscaling of monthly mean temperature data over a European region
Climate change is the most vital environmental change that has already started to affect many ecosystems. It is caused by greenhouse gas emissions which are increasing since the pre-industrial era, and populated areas become more vulnerable to disasters due to climate change. It has never been more crucial to model the climate effects on local regions. Organizations like Intergovernmental Panel on Climate Change (IPCC) use global climate models (GCMs) to project future changes in climate on a continental scale. Although these models are becoming more accurate, downscaling these models to smaller scales is an important task that is studied by climate scientists. The two main downscaling methods are dynamical and statistical downscaling. Statistical downscaling studies are more reachable and important to develop when compared to dynamical downscaling due to its lower costs. The use of machine learning algorithms in statistical downscaling is a new area. Studies that implement machine learning to make local scale projections of surface temperature are numbered. In this paper, four different machine learning algorithms were tested on downscaling of two different surface temperature datasets over a European region with different resolutions. The best performing algorithm was also tested augmenting elevation data. The results show that Gaussian process regression performs the best with MAE of 0.04 - 0.51 as compared to the other machine learning algorithms tested. In conclusion, machine learning algorithms such as Gaussian process regression can be a suitable approach when downscaling spatial monthly mean surface temperature data.
Predicting countries' vulnerability to climate change
Many countries are subject to the consequences of global climate change at varying degrees, and their vulnerability varies according to socioeconomic and environmental factors. Individuals, societies, and countries must be aware of the effects of climate change. Knowledge and understanding of how exposed they are to hazards, their level of vulnerability, and what must be done are critical for the continuation of basic vital activities. Accurately predicting how much a country will be affected by climate change in the future or which life-supporting sectors will suffer is crucial for countries to take precautions. Therefore, in this study, The Notre Dame Global Adaptation Initiative (ND-GAIN) Country Index's data is used for predicting countries' vulnerability to climate change. It's an open-source index that displays how vulnerable a nation is to climate disruptions. The data for this index includes scores for vulnerability and six areas that support life, including food, water, health, ecosystem services, human habitat, and infrastructure, for 182 nations during a 26-year period from 1995 to 2020. Long short-term memory (LSTM) network-based model is built to predict seven countries' six years of data from 2021 to 2026. These countries are Turkey, Australia, Germany, Portugal, Sudan, Georgia, and Kyrgyzstan. The results showed that the model predicts an increase in the vulnerability scores of all countries except Sudan for 2021, a slight decrease in Germany and Australia, and a decrease in Turkey, Portugal, and Kyrgyzstan after 2021. The model predicts a decrease in Sudan and an increase in Georgia for all years. The model's successes are tested using data from 2010 to 2020. Although time series forecasting is challenging, forecasted values are close to actual values. This study is novel since no other studies have predicted countries' future years' vulnerability to climate change.
Life cycle analysis of different cuisines
The extensive global food system is responsible for approximately 30% of greenhouse gas emissions. While the steps of the global food system such as production, packaging, transportation, distribution, storage, and disposal are followed, environmental effects remain intangible. With the life cycle assessment (LCA), environmental impacts are seen in concrete form at every stage of the system. In this study, the environmental impacts of different cuisines were investigated through life cycle assessment. Three menus have been created, consisting of Turkish, Far East, and Mediterranean cuisines, which are known and have a wide variety of food. Each menu has been chosen in accordance with the culture of the cuisine it has. The menus consist of soup, main course, side course and dessert. As a result of the life cycle assessment made on the menus selected for 3 cuisines, it has been determined that the environmental impact of the Mediterranean cuisine is quite low. The reason why the environmental impact is very low compared to the Turkish and Far Eastern cuisines, mainly agricultural foods are included in the Mediterranean cuisine and animal- based meals are not preferred much. On a food basis, the environmental impact of animal- based foods is greater than that of plant foods. As a result of the study, Turkish cuisine, in which animal-based meals are predominant, is the cuisine with the most environmental impact.
Interpretation of compound fragments via attentive recursive tree
The discovery of new drug-like chemicals with desired properties is a challenging and costly process in the pharmaceutical industry. To facilitate this process in the preclinical phase, many different neural network models have been proposed for different tasks (e.g., drug-target affinity prediction, molecular property prediction, target-specific molecule generation). Despite producing successful results, they usually lack interpretability. To comprehend the significance of each fragment in the relevant compounds, we employed the Attention Recursive Tree (AR-Tree) model. Thanks to its task-specific attention mechanism, AR-Tree highlights the significant fragments of compounds by positioning them closer to the root of the tree structure. In this way, the identified significant fragments can be used to design new compounds with desired properties in future research. We experimented with five different classification and four different regression tasks of the MoleculeNet as benchmark tasks. The results of the experiments show that the proposed architecture succeeded in finding chemically meaningful fragments for the corresponding tasks.
Convolutional ensemble learning for edge intelligence
Deep Edge Intelligence targets the deployment of deep learning algorithms in the edge network. While training deep networks requires computational resources, edge devices frequently lack high computational power. Decentralized learning methods such as federated learning provide a solution for gathering limited information from edge devices and collectively improving prediction performance. However, a drawback of such methods is that they often require multiple rounds of network communication, which increases communication time and the risk of communication errors. Another drawback is that the same model architecture is often used on all edge devices, which makes it mandatory to work with devices above a level of computational capacity. This thesis proposes a hybrid learning approach that employs ensemble learning with a convolutional scheme for different edge model architectures, except for a selected fully connected layer of the same dimensionality. Initially, shallow neural networks are trained on edge devices until a certain level of performance is achieved. Next, the feature representations obtained by the shallow models are transferred to an ensemble model. Subsequently, the proposed convolutional ensemble model is trained to boost the prediction performance. This method facilitates the completion of the system training with a one-way data transfer between edge devices and the server. Variational auto-encoders are also utilized to generate feature vectors in case transferring the required representations from the edge devices fails. Extensive experiments demonstrate that the suggested method outperforms state-of-the-art techniques in terms of accuracy while requiring fewer communications and a lower amount of data in various training scenarios.