Scopus İndeksli Yayınlar Koleksiyonu / Scopus Indexed Publications Collection
Permanent URI for this collectionhttps://hdl.handle.net/11147/7148
Browse
7 results
Search Results
Article Integrating QSAR Analysis and Machine Learning To Explore the Antidiabetic Potential of Natural Compounds(AMG Transcend Association, 2025) Sincar, B.; Yalcin, D.; Bayraktar, O.This study explores the antidiabetic potential of 72 natural compounds using molecular descriptors and QSAR modeling combined with machine learning techniques. The dataset includes 11 experimentally obtained compounds and 61 from the literature, characterized by their IC50 values indicating 50% inhibition of α-glucosidase enzyme activity. Molecular descriptors were generated using ChemAxon’s MarvinSketch and PADEL software, narrowing down over 3000 descriptors to 23 relevant features. Statistical analysis revealed significant multicollinearity among variables, necessitating the application of non-linear machine learning models, namely Random Forest and Gradient Boosting. These models demonstrated predictive capabilities with R² values of 0.7751 and 0.8066, respectively, and highlighted molecular weight and the number of heteroatoms in ring structures as critical features influencing IC50 values. Despite the dataset's variability and limited size, the study underscores the potential of integrating QSAR and machine learning approaches to effectively predict the antidiabetic activity of natural compounds. The findings provide valuable insights for advancing computational methods in drug discovery. © 2025 by the authors.Article Citation - WoS: 2Citation - Scopus: 1A Novel Framework for Droplet/Particle Size Distribution in Suspension Polymerization Using Physics-Informed Neural Network (PINN)(Elsevier Science Sa, 2025) Turan, Meltem; Dutta, AbhishekA Machine Learning (ML) based neural network can capture the complex evolution of polymer chain distributions, accounting for factors such as initiation, propagation, and termination steps in a suspension polymerization process, by integrating stagewise molar balance model (MBM) and population balance model (PBM) with Physics-Informed Neural Network (PINN). The integrated PINN framework is proposed to efficiently solve these equations, incorporating known physical laws as constraints and minimizing errors in both the distribution and dynamics of the polymer chains. By optimizing the neural network parameters such as weight matrices and bias vector, the model reproduces the moments of the polymer molecular weight distribution in close alignment with numerical solutions, and it generates population balance solutions that exhibit excellent agreement with their analytical counterparts. Sensitivity analyses for the depth of the neural network architecture to quantify how structural choices affect model fidelity has been performed. The resulting MBM-PINN and PBM-PINN integrated framework demonstrates robustness and versatility in accurately capturing (96-97%) droplet/particle dynamics. The proposed methodology has the capability to provide a powerful tool for faster and scalable simulations of polymerization reactions, enabling better prediction of product properties which could be used for optimizing reaction conditions in industrial applications.Article Technology-Enhanced Multimodal Learning Analytics in Higher Education: a Systematic Literature Review(Institute of Electrical and Electronics Engineers Inc., 2025) Raşıt Yürüm, O.Multimodal learning analytics (MMLA) is an emerging field of learning analytics and promises a more comprehensive analysis of the learning process thanks to advances in technological devices and data science. The purpose of this study was to explore technology-enhanced multimodal learning analytics in higher education systematically. A systematic literature review was performed using the PRISMA guidelines, and 45 studies published between January 2012 and June 2024 were determined. The findings demonstrated that China, the USA, Australia, and Chile were the leading contributors to MMLA research, with a notable surge in publications in 2021. Audio recorders, cameras, webcams, eye trackers, and wristbands were the most used devices. Most studies were conducted in experiment rooms or laboratories, though studies in authentic classroom settings have been growing. Data were primarily collected during activities such as programming, simulation exercises, presentations, discussions, writing, watching videos, reading, or exams, as well as throughout the entire instructional process, predominantly in computer science, health, and engineering courses. The studies were mainly predictive or descriptive whereas quite a few studies were prescriptive. Frequently tracked data types included audio, gaze, log, facial expression, physiological, and behavioral data. Traditional machine learning and basic statistics were the commonly used analytical methods whilst advanced statistics and deep learning were relatively less utilized. Test performance, engagement, emotional state, debugging performance, and learning experience were the popular target variables. The studies also pointed out several implications and future directions, with a significant portion highlighting the development of interventions, frameworks, or adaptive systems using MMLA. © 2013 IEEE.Article Predicting the Area Moment of Inertia of Beam and Column Using Machine Learning and Hypernetexplorer(Springer Science and Business Media Deutschland GmbH, 2025) Aydın, Y.; Nigdeli, S.M.; Roozbahan, M.; Bekdaş, G.; Işıkdağ, Ü.Beams and columns are the most important elements of steel frame structures. Damage to the beam or column can lead the structure to serious hazards and cause collapse. In the structural engineering literature, it has been observed that there is not much work for area moment of inertia estimation of beam and column. The aim of this study was to predict the area moment of inertia of beam and column using HyperNetExplorer developed by the authors. This method aims to bring innovation by optimizing artificial neural networks (ANNs). In this study, a prediction study is performed using 306 collected data on beam and column area moment of inertia. Classical ML models (linear regression (LR), decision tree regression (DTR), K neighbors regression (KNN), polynomial regression (PR), random forest regression (RFR), gradient boosting regression (GBR), histogram gradient boosting regression (HGBR)) and NAS and HyperNetExplorer were applied to predict beam and column area moment of inertia. The prediction performances were compared using different performance metrics (coefficient of determination (R2) and mean squared error (MSE)) and HyperNetExplorer developed by the authors showed the highest performance (R2 = 0.98, MSE = 246.88). Furthermore, SHapley additive explanations (SHAP) were used to explain the effects of features in the prediction models and it was observed that the most effective features for model predictions were loading on beam and length. The results show that the proposed NAS base approach and the developed tool, HyperNetExplorer, provides better performance when compared with classical ML methods. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2025.Article Citation - WoS: 3Citation - Scopus: 3Data Driven Modeling and Design of Cellulose Acetate-Polysulfone Blend Ultrafiltration Membranes Based on Artificial Neural Networks(Elsevier Ltd, 2025) Gungormus, E.This study aimed to develop and validate an Artificial Neural Networks (ANNs) model for the design and optimization of cellulose acetate-polysulfone blend ultrafiltration membranes, produced via the Non-Solvent Induced Phase Separation method. After some data science applications on a comprehensive dataset obtained from literature studies, the ultimate ANNs model exhibited superior predictive capabilities and effectively captured complex nonlinear relationships in the data. The optimum model configuration with a single hidden layer containing six neurons provided reliable predictions by avoiding overfitting and underfitting risks and significantly reducing error metrics. The model analyzed the effects of input variables on outputs, revealing that different stages of the membrane preparation process had varying impacts on performance metrics. This finding emphasized the importance of systematically optimizing the preparation process to enhance overall membrane performance. The model's predictions showed strong agreement with experimental data, further validating its accuracy. The optimum production conditions identified by the model offered significant improvements in membrane performance. Moreover, the model accelerated the membrane development process by reducing the required number of experimental trials and promoting efficient resource utilization. This approach contributed to both economic and environmental sustainability by reducing production costs and energy consumption. This study highlighted the significant potential of machine learning techniques for future innovations and advancements in this field by enabling precise, efficient, and sustainable membrane design and synthesis. © 2025 Elsevier Ltd.Conference Object Citation - WoS: 1Konteyner Görüntülerini Kullanarak Hasar Tespiti ve Sınıflandırması(IEEE, 2020) Imamoglu, Zeynep Ekici; Tuglular, Tugkan; Bastanlar, YalinIn the logistics sector, digital transformation is of great importance in terms of competition. In the present case, container warehouse entry / exit operations are carried out manually by the logistics personnel including container damage detection. During container warehouse entry / exit process, the process of detecting damaged containers is carried out by the personnel and several minutes are required to upload to the IT system. The aim of our work is to automate the detection of damaged containers. This way, the mistakes made by the personnel will be eliminated and the process will be accelerated. In this work, we propose to use a convolutional neural network (CNN) that takes the container images and classify them as damaged or undamaged. We modeled the problem as a binary classification and employed different CNN models. The result we obtained shows that there is no single best method for the classification. It is shown how the dataset was created and how the parameters used in the layered structures affect the models employed in this study.Article Citation - Scopus: 4Quasi-Supervised Strategies for Compound-Protein Interaction Prediction(John Wiley and Sons Inc, 2022) Çakı, O.; Karaçalı, B.In-silico compound-protein interaction prediction addresses prioritization of drug candidates for experimental biochemical validation because the wet-lab experiments are time-consuming, laborious and costly. Most machine learning methods proposed to that end approach this problem with supervised learning strategies in which known interactions are labeled as positive and the rest are labeled as negative. However, treating all unknown interactions as negative instances may lead to inaccuracies in real practice since some of the unknown interactions are bound to be positive interactions waiting to be identified as such. In this study, we propose to address this problem using the Quasi-Supervised Learning (QSL) algorithm. In this framework, potential interactions are predicted by estimating the overlap between a true positive dataset of compound-protein pairs with known interactions and an unknown dataset of all the remaining compound-protein pairs. The potential interactions are then identified as those in the unknown dataset that overlap with the interacting pairs in the true positive dataset in terms of the associated similarity structure. We also address the class-imbalance problem by modifying the conventional cost function of the QSL algorithm. Experimental results on GPCR and Nuclear Receptor datasets show that the proposed method can identify actual interactions from all possible combinations. © 2021 Wiley-VCH GmbH.