Bütünleme sınavları ara sınavlar ve genel sınavlar neticesinde dersi başaramayan öğrencilerin katıldığı bir sınavdır ve ülkemizde birçok üniversite tarafından uygulanmaktadır. Her ne kadar dersi başaramayan öğrencilerin bu sınavlara katılacağı varsayılsa da öğrencilerin bir kısmı çeşitli içsel (ara sınav notunun çok düşük olması, dersi öğrenme amacıyla yeniden alma isteği, tatil yapma ihtiyacı vb.) ve dışsal motivasyonlarla (memleketi ile üniversite arası mesafe, maddi imkânsızlıklar vb.) bütünleme sınavına katılmamaktadır. Tüm öğrencilerini sınava katılacağı varsayılarak yapılan sınav programları, gözetmen atama ve sınav kâğıdı çoğaltılması gibi süreçler ise maddî israflara neden olmaktadır. Bu çalışmada bütünleme sınavına kalan öğrencilerin bütünleme sınavına katılıp katılmayacağının tespit edilmesi amacıyla aşırı öğrenme makinesi tabanlı bir yaklaşım önerilmiştir. Önerilen yaklaşımda birden fazla aşırı öğrenme makinesi aynı veri setinin belirli bir oranındaki örnekler ile eğitime tabi tutulmuş ve test aşamasında oylama tabanlı bir yaklaşım ile sınıflandırma yapılmıştır. Oylamaya katılacak ELM karar parametrelerinin optimizasyonu ise yapay arı kolonisi algoritması ile sağlanmıştır. Önerilen yöntemlerin eğitimi ve test edilmesi amacıyla ise Konya Teknik Üniversitesi Bilgisayar Mühendisliği Bölümü’nde okutulan 9 adet derse ait veriler toplanmıştır. Ayrıca 9 adet dersin birleşiminden oluşan bir veri kümesi daha oluşturulmuş ve önerilen yöntemin analizi bu 10 veri seti üzerinde yapılmıştır. Deneysel çalışmalarda aşırı öğrenme makinesi ve önerilen yöntemin performansı her ders için ayrı ayrı karşılaştırılmıştır. Önerilen yöntemin performansının sınıflandırma doğruluğu açısından temel aşırı öğrenme makinesinden daha yüksek olduğu gözlemlenmiştir

This study introduces two hybrid forecasting models that integrate the Marine Predators Algorithm (MPA) with Adaptive Neuro-Fuzzy Inference Systems (ANFIS) and Feed-Forward Neural Networks (FFNN) for short-term Bitcoin price prediction. Daily Bitcoin data from 2022 were converted into supervised time-series structures with multiple input configurations. The proposed hybrid models were evaluated against six well-known metaheuristic algorithms commonly used for training intelligent forecasting systems. The results show that MPA consistently yields lower prediction errors, faster convergence, and more stable optimization behavior compared with alternative algorithms. Both ANFIS-MPA and FFNN-MPA maintained their advantage across all tested structures, demonstrating reliable performance under varying model complexities. All experiments were repeated multiple times, and the hybrid approaches exhibited low variance, indicating robust and reproducible behavior. Overall, the findings highlight the effectiveness of MPA as an optimizer for improving the predictive performance of neuro-fuzzy and neural network models in financial time-series forecasting.

Metaheuristic optimization algorithms have become one of the most important research areas for solving complex, nonlinear, and high-dimensional optimization problems. These algorithms are widely used in many application domains such as engineering, artificial intelligence, energy systems, healthcare, communication technologies, manufacturing, data science, and industrial applications. Their strong exploration and exploitation capabilities enable them to provide effective solutions for optimization problems where classical optimization methods are insufficient. In addition, many metaheuristic optimization algorithms attempt to establish a balanced and adaptive search behavior between exploration and exploitation processes, reflecting an important form of search symmetry during the optimization process. This systematic review presents a comprehensive analysis of metaheuristic optimization algorithms based on 515 distinct metaheuristic optimization algorithms. In this study, the algorithms are systematically classified into eleven major categories according to their inspiration sources and dominant search behaviors: Evolutionary-Based, Swarm Intelligence-Based, Bio-Inspired, Ecological/Environmental-Based, Physics-Based, Mathematics-Based, Human-Based, Game/Sport-Based, Chemical and Molecular-Based, Music-, Art-, and Culture-Inspired, and Hybrid and Advanced Metaheuristics. The review analyzes the historical evolution, inspiration diversity, search structures, symmetry-oriented search balance, and development trends of modern metaheuristic optimization algorithms. This study also provides a broader understanding of how metaheuristic optimization algorithms have evolved structurally and conceptually over time. In addition, the study highlights the importance of balanced and adaptive search mechanisms for maintaining optimization efficiency under complex problem conditions. The findings show that the number and diversity of metaheuristic optimization algorithms have increased significantly, especially after 2015. Population-based search structures remain the dominant optimization approach in the literature, while hybrid optimization methods, adaptive search mechanisms, and artificial intelligence integration have become major research trends in recent years. The study also discusses important challenges in the field, including algorithm similarity, benchmark standardization, theoretical limitations, and fair performance evaluation.

Crude oil is one of the most important assets that are used in the production of many industrial products in a wide variety of areas. The importance of crude oil has made it important to predict its future price. Therefore, it is possible to come across many studies in the literature in which the price of crude oil is estimated in the short or long term. In this study, innovative adaptive neuro-fuzzy inference systems (ANFIS) based approaches are proposed to estimate the daily minimum and maximum prices of crude oil. The used data was taken from the period between January 3, 2022, and December 29, 2023. A total of 516 different days of data were collected to create the dataset for analysis. For daily forecasting, time series data were transformed into a data set consisting of two inputs and one output. Moth-flame optimization algorithm (MFO), flower pollination algorithm (FPA), biogeography-based optimization (BBO) and artificial bee colony (ABC) were used in training ANFIS. The results obtained in the training and testing processes were compared. When the results obtained were compared, it was shown that the relevant algorithms were effective in the daily estimation of crude oil. It has been observed that effective results are also achieved at low evaluation numbers, especially thanks to the fast convergence feature of the MFO and BBO algorithms.

This study concentrates on the weighted average ensemble-based prediction of pIC50 value for Human Dihydroorotate Dehydrogenase (hDHODH) using hybrid molecular fingerprints. By querying the ChEMBL database for IC50 data, a diverse collection of 1585 molecules was obtained, and these values were converted to pIC50 values to develop ensemble-based prediction models. We used the weighted average (W.Avg) of Light Gradient Boosting Machine (LGBM), Bootstrap Aggregating (Bagging), and Random Forest (RF) algorithms to estimate pIC50 values. Model performance was evaluated using 5x3 repeated K-fold cross-validation (CV). Root mean square error (RMSE) and mean squared error (MSE) were used as the performance metrics. The W.Avg combination demonstrated overall success beyond individual models. The results showed that our ensemble model outperformed all other baseline models with R²=0.8266, RMSE=0.6568, and MSE=0.4337. Paired t-test results indicate that the W.Avg model is statistically significantly superior to the other models in terms of R², RMSE, and MSE (p < 0.05). This ensemble-based method accelerated hDHODH inhibitor discovery by reducing screening time and increasing predictive accuracy.