Scopus İndeksli Yayınlar Koleksiyonu / Scopus Indexed Publications Collection
Permanent URI for this collectionhttps://hdl.handle.net/11147/7148
Browse
18 results
Search Results
Conference Object A General Predictive Model to Evaluate Daylight Levels of Residential Buildings in the Mediterranean (Next Med) Region(Education and Research in Computer Aided Architectural Design in Europe, 2025) Ekici, B.Conceptual design is one of the most critical phases, as design decisions affect the buildings’ performance throughout their life cycle. Researchers consider various computational methods to achieve effective design proposals. Nevertheless, optimization algorithms are necessary to cope with the complexity and increase the efficiency of design alternatives in various aspects. In sustainable building design, these decisions require computationally expensive processes due to the simulation tasks. Besides, making sustainable design decisions is even more challenging in a Mediterranean climate due to changing conditions throughout the year. Therefore, recent studies frequently consider combining predictive models with optimization algorithms to decrease the burden of expensive simulation time. Relevant works present promising outcomes, yet they are limited to predicting the building performance of specific cases; thus, the proposed predictive models are limited to different design problems. This paper investigates the development of a general machine learning (ML) model to overcome this issue. With this motivation, a parametric test box consisting of twenty parameters related to weather data of twelve Mediterranean (Next Med) countries, space dimensions, vertical/horizontal louvers, and material type is developed using Grasshopper 3d. Moreover, a parametric urban model, which considers eight parameters related to the density of the surrounding buildings, is also created to generate numerous environments. The LadyBug tools simulate the daylight autonomy to generate 12,000 samples. Five different ML models involving artificial neural networks (ANN) are built in Python. Statistical results showed that train and test scores achieved promising outcomes in all ML models. However, when predicting user-defined scenarios not involved in the generated dataset, only ANNs perform generalizable, accurate predictions. The paper discusses the ability of ANN models to accurately predict different design scenarios and locations, and the trustworthiness of the training and test scores based only on collected data. © 2025, Education and research in Computer Aided Architectural Design in Europe. All rights reserved.Article Artificial Intelligence for Improving Thermal Comfort through Envelope Design in Residential Buildings: Recent Developments and Future Directions(Elsevier Science Sa, 2026) Bayraktar, Arda; Ekici, BerkEnvelopes are vital components for improving thermal comfort in almost all building typologies. Yet, the design and analysis of envelopes are complex, as they involve multiple aspects and various parameters, ensuring comfort standards. Improving thermal comfort in residential buildings is within the scope of researchers to suggest sustainable design alternatives that consider multiple performance aspects and design parameters. Previous review articles have focused on improving thermal performance in residential buildings from the perspective of envelope technology, materials, and design strategies. However, none of them investigated current developments using artificial intelligence (AI), which inevitably supports decision-making in complex circumstances for a sustainable built environment. This review examines the contribution of AI methods, which consist of metaheuristic optimization and machine learning algorithms as sub-branches, to envelope parameters. The paper systematically reviews 95 relevant works on AI, including early approaches, to provide a comprehensive overview of current developments, following PRISMA guidelines. The results showed that early applications considered conventional approaches to improve thermal comfort and energy performance, which mostly limit the results to specified cases. On the other hand, studies utilizing AI methods dealt with numerous parameters, allowing them to cope with complex envelope systems in a reasonable amount of time. The study addresses relevant research questions related to the trends, research methods, system types, AI methods, data types, and their relation to performance and envelope parameters. The study also provides actionable insight, underlining gaps and future works for utilizing machine learning methods in the reviewed research domain.Conference Object Developing Machine Learning Models to Predict Outdoor Thermal Comfort of Kinetic Shading Devices: An Approach for Global Optimization(Education and Research in Computer Aided Architectural Design in Europe, 2025) Dağlier, Y.; Ekici, B.; Korkmaz, K.Utilizing artificial intelligence (AI) methods in the design process supports the achievement of sustainable alternatives during the conceptual design. In various AI methods, optimization and machine learning (ML) algorithms are the most common methods to develop predictive models and discover favorable design alternatives with significantly reduced computational time. Recent works focused on limited datasets, as well as the evaluation of the developed prediction models based on collected data. During the optimization process of complex design problems, the number of design parameters becomes enormous; thus, search areas contain many design alternatives that might lead the search outside of the collected data. Therefore, evaluating the accuracy of prediction models only based on the collected samples may result in scenarios where the predicted outcome during the optimization process aligns with an unrealistic solution. This study investigates how accurately prediction models developed using different ML algorithms can perform in optimization processes. The proposed framework is used to cope with outdoor thermal performance, considering kinetic shading devices with rigid origami techniques. A parametric shading device model with kinematic principles and 10 design parameters is created in Grasshopper 3d. LadyBug is used to analyze the performance of the universal thermal climate index (UTCI). To minimize the UTCI, the radial basis function optimization (RBFOpt) algorithm in the Opossum plugin is used. To compare the optimization results with the prediction results, multiple linear regression, support vector machines, random forest, polynomial regression algorithms, and artificial neural networks (ANN) are developed to predict outdoor thermal comfort performance targets on each collected data set with 2000 samples. Results showed that ANN models can provide more accurate predictions during the optimization process. The paper aims to discuss the way ML algorithms are applied and evaluated for ML-based optimization domains in design problems. © 2025, Education and research in Computer Aided Architectural Design in Europe. All rights reserved.Article Development and Validation of Regression Model via Machine Learning to Estimate Thermal Conductivity and Heat Flow Using Igneous Rocks from the Dikili-Bergama Geothermal Region, Western Anatolia(Pergamon-Elsevier Science Ltd, 2026) Ayzit, Tolga; Sahin, Onur Gungor; Erol, Selcuk; Baba, AlperThermal conductivity is a fundamental parameter that significantly influences the thermal regime of the lithosphere. It plays a crucial role in a variety of geological applications, including geothermal energy exploration, igneous system assessment, and tectonic modeling. In this study, a machine learning approach is used to predict the thermal conductivity of igneous rocks based on the composition of major oxides. A total of 488 samples from different regions of the world were analyzed. The thermal conductivity values ranged from 1.20 to 3.74 Wm(-1) K-1 and the mean value was 2.61 Wm(-1) K-1. The Random Forest (RF) algorithm was used, resulting in a high coefficient of determination (R-2 = 0.913 for training and R-2 = 0.794 for testing) and a root mean square error (RMSE) of 0.112 and 0.179, respectively. Significance analysis of the traits identified SiO2 (>40 %), Na2O (>15 %) and Al2O3 (>10 %) as the most influential predictors. The study presented results from the Western Anatolia region, where felsic rocks had the highest thermal conductivity (mean = 2.69 Wm(-)(1)K(-)(1)) compared to mafic (mean = 2.34 Wm(-)(1)K(-)(1)) and ultramafic rocks (mean = 2.39 Wm(-)(1)K(-)(1)). In addition, the study evaluated the predictive capabilities of machine learning models for the igneous rocks of the Dikili-Bergama region and compared the results with those of saturated models. Using these data, we calculated heat flow values of up to 400 mWm(-2) under saturated conditions in western Anatolia. These results highlight the value of integrating geochemical data with machine learning to improve geothermal resource exploration and lithospheric modeling.Article A Knowledge-Driven Computer Vision Framework for Automated Atomic Force Microscopy Surface Characterization(Elsevier Science Ltd, 2026) Deveci, D. Gemici; Barandir, T. Karakoyun; Unverdi, O.; Celebi, C.This study presents an innovative analytical framework developed to automate Atomic Force Microscopy (AFM)-based surface characterization. The proposed methodology integrates computer vision (CV) algorithms and machine learning (ML) techniques to overcome the limitations of conventional observer-dependent approaches and visual inspection methods. In the first stage of the two-step data processing pipeline, raw AFM signals were converted into structured datasets, correspondences between images acquired under different loading conditions were identified, and drift effects in both direction and magnitude were predicted using a LightGBM-based machine learning (ML) model to guide subsequent analytical processes. This process establishes a unified coordinate reference across varying force levels, enabling pixel-level comparability of surface maps. In the second stage, the aligned datasets are systematically analyzed through block-based local maxima detection, edge-based contour extraction, morphological filtering, and skeletonization algorithms. In this way, ridge-like surface features are reliably identified and quantitatively evaluated along their axes under varying force conditions. The framework ensures data integrity while enabling high-resolution and reproducible analyzes. Beyond its automation capability, it is distinguished by its integrated, modular architecture, where each component operates sequentially along a unified processing pipeline. The methodology was validated using epitaxial monolayer graphene grown on the C-face of SiC, successfully demonstrating its ability to resolve both geometric and force-dependent mechanical responses. In this regard, the proposed system extends beyond conventional cross-sectional analysis by providing a drift-aware, knowledge-guided compensation mechanism and directionally resolved evaluation, offering a robust, automation-ready infrastructure for nanoscale surface characterization.Article Knowledge-Based Training of Learning Architectures Under Input Sensitivity Constraints for Improved Explainability(Pergamon-Elsevier Science Ltd, 2026) Sildir, Hasan; Erturk, Emrullah; Edizer, Deniz Tuna; Deliismail, Ozgun; Durna, Yusuf Muhammed; Hamit, BahtiyarThe traditional machine learning (ML) training problem is unconstrained and lacks an explicit formulation of the underlying driving phenomena. Such a formulation, based solely on experimental data, does not ensure the delivery of qualitative knowledge among variables due to many theoretical issues in the optimization task. This study further tightens Artificial Neural Networks (ANNs) training by including input sensitivities as additional constraints and applies to regression and classification tasks based on literature data. In theory, such sensitivity represents the change direction of the target variable per change in measurements from indicators. The resulting nonlinear optimization problem is solved th rough a rigorous solver and includes the sensitivity expressions through algorithmic differentiation. Compared to traditional methods, with an acceptable decrease in the prediction capability, the proposed model delivers more intuitive, explainable, and experimentally verifiable predictions under input variable variations, under robustness to overfitting, while serving robust identification tasks. A classification case study includes a patient-oriented clinical decision support system development based on the impact of cancer-indicating variables. A competitive test prediction accuracy is obtained compared to commonly used algorithms despite 10 % decrease in the training. The regression case is built upon the energy load estimation to account for prominent considerations to obtain desired sensitivity patterns and proposed methodology delivers significant accuracy drop compared to some formulations to address knowledge patterns. The approach delivers a compatible pattern with practitioner expertise and is compared to widely used machine learning algorithms, whose performances are evaluated through common statistics in addition to multi-variable response graphs.Article Unveiling the Design Rules for Tunable Emission in Graphene Quantum Dots: A High-Throughput TDDFT and Machine Learning Perspective(Indian Acad Sciences, 2025) Ozonder, Sener; Ozdemir, Mustafa Coskun; Unlu, CanerThe ability to tailor the optical properties of graphene quantum dots (GQDs) is critical for their application in optoelectronics, bioimaging and sensing. However, a comprehensive understanding of how shape, size and doping influence their emission properties remains elusive. In this study, we conduct a systematic high-throughput time-dependent density functional theory (TDDFT) and machine learning analysis of 284 distinct GQDs, varying in shape (square, hexagonal, amorphous), size (similar to\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sim$$\end{document}1-2 nm) and doping configurations with elements B, N, O, S and P at varying concentrations (1.5-7%). Our findings reveal clear design principles for tuning emission wavelengths based on dopant type, concentration and GQD geometry. Notably, sulfur doping at specific concentrations consistently results in higher emission energies, with certain configurations yielding emissions within the visible range. By elucidating how quantum confinement effects, symmetry breaking and dopant-induced modifications govern GQD optical properties, we provide practical design rules for tailoring emission spectra for next-generation optoelectronic, bioimaging and sensing applications.Conference Object Machine Learning-Based Antenna Selection and Secrecy Capacity Analysis(Institute of Electrical and Electronics Engineers Inc., 2025) Erdurak, Burak; Erdoǧan, Eylem; Gürkan, FilizThe performance of machine learning methods was analyzed to optimize antenna selection in wireless communication systems, and system's secrecy performance was observed. To enhance the antenna selection process, Deep Neural Networks (DNNs), Convolutional Neural Networks (CNNs), and the KNearest Neighbors (KNN) algorithm were utilized. Channel vectors were used as model inputs, aiming to select the most optimal transmission path among N possible candidates. During the training phase, the antenna with the highest Signal-to-Noise Ratio (SNR) was selected for data labeling. The performance of Single-Input Multiple-Output (SIMO), Multiple-Input SingleOutput (MISO), and Multiple-Input Multiple-Output (MIMO) system architectures was evaluated using model accuracy and the F1-score. Additionally, the secrecy capacity corresponding to the selected antennas was computed, demonstrating the feasibility of secure communication. The results indicate that deep learningbased methods achieved higher accuracy, with the CNN model emerging as the most successful approach, reaching an accuracy of over 95% across all system configurations. © 2025 Elsevier B.V., All rights reserved.Article An Alternative Software Benchmarking Dataset: Effort Estimation With Machine Learning(Elsevier Science Inc, 2026) Yurum, Ozan Rasit; Unlu, Huseyin; Demirors, OnurEffort estimation plays a vital role in software project planning, as accurate estimates of required human resources are essential for success. Traditional estimation models often depend on historical size and effort data, yet organizations frequently struggle to access reliable effort records. Public benchmarking datasets like ISBSG offer useful data but may lack coverage or involve licensing fees. To address this issue, we previously introduced a free, extendable benchmarking dataset that integrates functional size and effort data extracted from 18 studies. In this study, we examine the effectiveness of our dataset for predictive effort estimation and compare it with the widely used ISBSG dataset. Our analysis includes 337 records from our dataset and 732 ISBSG projects, focusing on those with COSMIC size data. We first developed and compared models using linear regression and nine machine learning algorithms - Bayesian Ridge, Ridge Regression, Decision Tree, Random Forest, XGBoost, LightGBM, k-Nearest Neighbors, Multi-Layer Perceptron, and Support Vector Regression. Then, we selected the best-performing models and applied them to an unseen evaluation dataset to assess their generalization performance. The results show that machine learning performance varies based on evaluation method and dataset characteristics. Despite having fewer records, our dataset enabled more accurate predictions than ISBSG in most cases, highlighting its potential for effort estimation. This study demonstrates the viability of our dataset for building predictive models and supports the use of machine learning in improving estimation accuracy. Expanding this dataset could offer a valuable, open-access resource for organizations seeking effective and lowcost estimation solutions.Article Comprehensive Analysis and Machine Learning-Based Solutions for Drift Behavior in Ambient Atomic Force Microscope Conditions(Pergamon-Elsevier Science Ltd, 2025) Deveci, D. Gemici; Barandir, T. Karakoyun; Unverdi, O.; Celebi, C.; Temur, L. O.; Atilla, D. C.This study outlines the effectiveness of combining numerical methods, Computer Vision (CV) and Machine Learning (ML) approaches to analyze and predict drift behavior in high-resolution Atomic Force Microscope (AFM) scanning procedures. Using Long Short-Term Memory (LSTM) models for time series analysis and the Light Gradient Boosting Machine (LightGBM) algorithm for predictive modeling, significant progress was achieved in understanding the dynamic and variable nature of drift and mitigating its impact on scanning. The models demonstrated a robust predictive capability, achieving approximately 94% accuracy in drift predictions. The study emphasizes the nonstationary characteristics of drift and demonstrates how the selection of features directly related to the target variable enhances the efficiency of the model and enables adaptive real-time correction. These findings confirm the predictive strength of the models and highlight the potential for integrating ML predictions with real-time feedback mechanisms to improve the resolution and stability of AFM imaging in both scientific and industrial applications.