Osman Altay

The performance of Multi-Layer Perceptron (MLP) models depends on the effective optimization of weight and bias parameters during the training process. Gradient-based learning methods may not always produce stable and high-performance results due to factors such as initial sensitivity, the need for derivative information, and stagnation at local minima in complex, high-dimensional search spaces. In this study, four current metaheuristic algorithms for optimizing the weight and bias parameters of MLPs were evaluated comparatively: Hippopotamus Optimization Algorithm (HO), Puma Optimizer, Starfish Optimization Algorithm (SFOA), and War Strategy Optimization Algorithm (WSO). The algorithms were tested on the Diabetes, Diagnosis, Mammographic Mass, and Wisconsin Breast Cancer datasets. The same MLP architecture and experimental conditions were used for each dataset; the data were evaluated with an 80% training and 20% testing split. Each method was run in 10 independent runs, with a population size of 30 and a number of iterations of 100, and candidate solutions were limited to the range [-10, 10]. Performance was measured by accuracy, precision, and F1-score; the overall ranking of the methods was examined using the Friedman test. The findings show that metaheuristic performance is sensitive to the dataset. In general, Puma-based training produced more competitive results in accuracy and F1-scores in most datasets, while HO excelled in precision and stability in some scenarios. Although SFOA and WSO produced competitive results in certain datasets, they did not exhibit consistent superiority across all metrics. The results demonstrate that the selection of meta-heuristic methods for MLP training should consider dataset characteristics and target metrics and provide a practical comparison framework for medical decision support applications. In addition, mean and standard deviation values are reported for each dataset, and the strengths of the methods are discussed based on the best run results.

Software data is widely used to predict attributes of software systems; however, obtaining reliable datasets from commercial companies remains challenging due to confidentiality constraints. GitHub has emerged as a data source, offering access to diverse applications and development statistics. Nevertheless, concerns about the reliability and representativeness of public repositories persist. Star count is a widely accepted indicator of repository popularity, and existing studies mainly rely on time-dependent platform metrics. In this study, we propose using static software metrics extracted from source code, along with GitHub statistics. To our knowledge, this study is among the first to use ELM for popularity prediction with static software metrics. Repositories from different application domains are selected to ensure dataset diversity and representativeness. An automated tool has been developed to collect data via the GitHub API and SourceMonitor CLI. In addition, several baseline machine learning models, including LR, SVM, RF, and LSBoost, are evaluated for comparison. Experimental results show that ELM achieves competitive performance across datasets. In terms of R² scores, ELM performs best in four datasets, RF in three, and LR in one. These results indicate that ELM is an effective method for popularity prediction and highlight the potential of incorporating static software metrics into GitHub-based predictive modeling.

The growing demand and power of modern power systems necessitates an economical and stable operation, and therefore, the Optimal Power Flow (OPF) problem is of heavy research interest. The OPF must be solved efficiently to lower operational costs and ensure system stability. A comprehensive comparative study in this paper compares the performance of five metaheuristic algorithms for solving the fuel cost minimization problem of the OPF problem. Two well-known algorithms, Particle Swarm Optimization (PSO) and Genetic Algorithm (GA), and three new ones, Dream Optimization Algorithm (DOA), Newton-Raphson-Based Optimizer (NRBO), and Greylag Goose Optimization (GGO), were tested on the benchmark IEEE 6-bus, 30-bus, and 57-bus systems. The performance of the algorithms was compared in terms of statistical measures of the best cost, mean cost, and standard deviation from ten independent runs. Numerical findings indicate that the performance of algorithms is highly dependent on system size. For the 6-bus case, NRBO had the lowest fuel cost, while PSO had better stability with the lowest standard deviation. In the 30-bus case, NRBO was the most effective algorithm to utilize, having better performance in all aspects measured. In the larger 57-bus system, PSO had the best solution overall but least consistent in performance. This study concludes that there isn't a single best algorithm for all OPF problem sizes, providing an authoritative benchmark that marks the strengths and weaknesses of classical and contemporary metaheuristics across different power system applications.

Objectives: The aim of this study was to comparatively evaluate four large language models (LLMs) used for patient education in ophthalmology in terms of accuracy, reliability, and patient safety across different ophthalmic subspecialties. Methods: In this cross-sectional evaluation, a total of 50 frequently asked patient questions covering five ophthalmic subspecialties (strabismus/pediatric ophthalmology, oculoplastics, cataract and refractive surgery, retina, and dry eye) were included. All questions were submitted in a text-only format to ChatGPT o3 Mini High, Gemini 2.0 Pro, Claude-Sonnet 3.7, and LLaMA 3.1 405B. The generated responses were independently evaluated by five blinded ophthalmologists using a 10-point scale assessing accuracy, currency, informativeness/clarity, and patient safety. Potentially unsafe content was identified and categorized using a predefined structured error taxonomy. Results: Marked differences in performance were observed among the models. Mean scores were 3.44 for Gemini, 2.99 for ChatGPT, 2.48 for Claude, and 1.09 for LLaMA. Gemini demonstrated higher performance across most subspecialties, whereas in the retina subspecialty, ChatGPT and Claude generated comparatively stronger responses. Of the 200 evaluated responses, 19 (9.5%) contained potentially unsafe content, with the lowest proportion observed for Gemini and the highest for LLaMA. Conclusions: LLMs can generate useful responses for patient education in ophthalmology, but performance varies by model and subspecialty. Within this 50-question, text-only expert-rating framework, Gemini 2.0 Pro and ChatGPT o3 Mini High provided relatively higher accuracy and reliability in most areas, whereas LLaMA 3.1 405B lagged. Larger and clinically integrated evaluations, including direct assessment of patient understanding and behavior, are needed to define their safe use in practice.

Biomedicine datasets often contain noisy, imbalanced, and high-dimensional features, which makes the choice of neural network training strategy a decisive factor for clinical decision support. While artificial neural networks (ANNs) have been widely applied to medical diagnosis tasks, comparative analyses across a broad spectrum of training algorithms remain limited. This study addresses this gap by evaluating twelve distinct ANN training algorithms on sixteen publicly available biomedicine datasets. The performance of each method was assessed according to training duration, loss metrics, and classification accuracy. During the testing phase, essential metrics such as F-measure, G-mean, accuracy, sensitivity, specificity, and precision were evaluated. A Friedman test was performed to statistically verify the performance discrepancies among the algorithms. The findings indicated that the selection of the training algorithm significantly influences classification efficacy. The results reveal that Levenberg-Marquardt and Bayesian Regularization consistently achieve high predictive performance on complex datasets, whereas the Variable Learning Rate method demonstrates robust and scalable performance across heterogeneous data conditions. Importantly, the findings indicate that no universal “best” algorithm exists; optimal selection depends on dataset characteristics such as size, dimensionality, and noise distribution. By mapping algorithm strengths and weaknesses across diverse diagnostic scenarios, this work provides actionable guidelines for researchers and practitioners aiming to design reliable and efficient ANN-based decision support systems in healthcare. Compared to baseline gradient descent variants, advanced training algorithms such as Levenberg-Marquardt and Bayesian Regularization achieved up to 25% higher accuracy (e.g., on the Skin_Nonskin dataset) and 20% higher F-measure (e.g., on the Saheart dataset) across multiple biomedicine datasets.

The purpose of this research is to use regression analysis to predict a CPU's performance score based on its features. CPU performance is incredibly important to evaluate when choosing a computer, along with system configuration and design. Support Vector Regression (SVR), Random Forest Regression (RFR), Multiple Linear Regression (MLR), Gradient Boosting Regression (GBR) and Neural Network Regression (NNR) are used to estimate the CPU's performance score. To test the algorithms, 30 percent of the data set was selected as test data and 70 percent as training data, separated randomly. As a result, the NNR has the highest of the coefficient of determination score which is 0.976, followed by GBR, 0.958. MLR, RFR and SVR algorithms have the R-squared score of 0.952, 0.934 and 0.865, respectively.

To ascertain the optimal and most efficient reservoir temperature of a geothermal source, long-term field studies and analyses utilizing specialized devices are essential. Although these requirements increase project costs and induce delays, utilizing machine learning techniques based on hydrogeochemical data can minimize losses by accurately predicting reservoir temperatures. In recent years, applying hybrid methods to real-world challenges has become increasingly prevalent over traditional machine learning methodologies. This study introduces a novel machine learning approach, named AOSMA-MLP, integrating the adaptive opposition slime mould algorithm (AOSMA) and multilayer perceptron (MLP) techniques, specifically designed for predicting the reservoir temperature of geothermal resources. Additionally, this work compares the basic artificial neural network and widely recognized algorithms in the literature, such as the whale optimization algorithm, ant lion algorithm, and SMA, under equal conditions using various evaluation regression metrics. The results demonstrated that AOSMA-MLP outperforms basic MLP and other metaheuristic-based MLPs, with the AOSMA-trained MLP achieving the highest performance, indicated by an R² value of 0.8514. The proposed AOSMA-MLP approach shows significant potential for yielding effective outcomes in various regression problems.