Halil Yiğit

Ensuring safe energy storage in lithium-ion batteries requires reliable fault diagnosis. However, classifying faults from small experimental datasets demands rigorous control over partition variability, hyperparameter bias, and measurement noise. This study evaluates five machine-learning classifiers using 110 experimentally measured charging observations (63 faulty, 47 non-faulty), each described by nine electrical and thermal features. The evaluation framework incorporated 30 repeated, stratified 80/10/10 partitions with an inner 5-fold cross-validation for hyperparameter tuning. LogitBoost showed the highest observed repeat-level mean performance, with an F1-score of 0.9930 ± 0.0214 (95% CI: 0.9841–1.0000) and accuracy of 0.9909 ± 0.0277 (95% CI: 0.9818–1.0000), closely followed by neural network and random forest models. The Friedman test revealed no statistically significant overall differences in performance among the models (chi-square = 7.6467, p = 0.1054). Under a computational zero-mean Gaussian perturbation test with amplitudes up to 10% of the development-set feature standard deviation, the mean F1-score drop remained below 0.0034, and prediction consistency exceeded 0.9881. Consensus feature rankings identified intermediate charging currents as most decisive, showing moderate stability across partitions (mean rank correlation = 0.6361). Although physical cell identifiers were not retained, these results demonstrate robust sample-level, within-dataset fault discrimination, providing a methodology for reliable evaluation of limited experimental battery data.

In the rapidly evolving landscape of the hospitality industry, comprehending and enhancing customer experience through digital feedback has become a strategic imperative. This study introduces a high-precision framework powered by advanced transformer-based architectures to evaluate customer sentiments in hotel reviews. A significant challenge in this domain, class imbalance, was addressed through an innovative data augmentation strategy utilizing ChatGPT to generate semantically consistent paraphrased samples for minority classes. We conducted a comprehensive comparative analysis of five state-of-the-art models: BERT, DistilBERT, RoBERTa, ELECTRA, and GPT-2. The methodology was validated using a specialized dataset, the Bali Hotel Review Corpus (BHRC), alongside four globally recognized benchmarks (IMDb, SST-1, SST-2, and MR). Experimental results demonstrate that the proposed framework achieves exceptional accuracy levels, with the ELECTRA model outperforming others by reaching a peak accuracy of 99.96% under balanced conditions. This performance is attributed to ELECTRA’s efficient discriminative pre-training objective, which proves superior to standard masked or generative approaches for domain-specific sequence classification. Our findings provide a robust and resource-efficient benchmark for real-time customer experience appraisal, offering hotel managers actionable insights to drive service quality in the modern digital tourism economy.

This paper presents a comprehensive evaluation of phishing detection architectures, systematically transitioning from single-modality text classifiers to parameter-efficient hybrid multimodal systems. Using the MTLP dataset, a large-scale collection containing 100,000 legitimate and phishing webpage samples, we systematically evaluated multiple model configurations. The dataset includes Uniform Resource Locators (URLs), HyperText Markup Language (HTML) content, and WHOIS metadata. Each experiment targeted a specific research question, examining the impact of Parameter-Efficient Fine-Tuning (PEFT) using Low Rank Adaptation (LoRA), feature fusion strategies, and feature engineering on detection accuracy and efficiency. The optimal configuration (V5.2) integrated a language-specific Turkish BERT encoder with 14 domain-specific URL features, achieving an F1-score of 99.14% and outperforming text-only, image-based, and WHOIS-based baselines. Our experiments show that increasing architectural complexity (e.g., cross-attention) does not necessarily lead to better performance; instead, a balanced fusion of linguistic and structural cues yields optimal results. We also observed a clear trade-off between parameter count and accuracy, as lightweight models with only 1,538 trainable parameters achieved near state-of-the-art performance. Error analysis revealed that even advanced systems struggle with structurally ambiguous or content-sparse webpages. Overall, the findings highlight the potential of domain-adapted hybrid architectures as a robust, efficient, and explainable approach to phishing detection.

This study focuses on cold deformation and age effects on the microhardness and electric conductivity of the Cu-4Ti (wt.%) alloys. The samples were solution treated at 900 °C, quenched in water, and aged at 450–600 °C for 1–120 min. Fifty percent cold rolling was performed before aging to analyze the impact on their microstructure and properties. Hardness and electric conductivity were examined by the Vickers microhardness and Förster testing. Hardness increased significantly while electric conductivity was maintained. The optimal hardness of 298 HV appeared following 50% cold rolling and aging for 120 min at 450 °C, and an electric conductivity of 9.4 MS/m was achieved after 120 min at 600 °C in cold-rolled materials. The deformed and solution-treated materials reached 244 HV after 120 min at 500 °C, and electric conductivity reached 7.7 MS/m. Polynomial models of regression were used to analyze the impact of aging parameters on properties. Process parameters were properly optimized by applying metaheuristic algorithms. These contributions ensure a better understanding of the relationship between the microstructure and properties in Cu-Ti alloys, as well as their application in aircraft and electronics.

The optimization of CuNi₂Si₁ alloy’s mechanical and electrical properties was achieved through a combination of experimental approaches and metaheuristic algorithms. Optimizing hardness and electrical conductivity through a variation in aging temperature (450–600 °C) and aging duration (1–420 min) was taken under consideration in the present work. Cold rolling with 50% strain after solution annealing aided in microstructure refinement and accelerated Ni₂Si precipitates’ development, and property improvement increased. Optimum temperature and holding period were 450 °C and 30 min, respectively, with 266 HV and 13 MS/m and 167 HV and 11.2 MS/m for non-deformed samples, respectively. SPBO, genetic algorithm (GA), and particle swarm optimization (PSO) metaheuristic algorithms were considered, and SPBO exhibited the best prediction accuracy. SPBO predicted 450 °C for 61.75 min, and experimental testing exhibited 267 HV and 14 MS/m, respectively. Polynomial regressions with 0.98 and 0.96 values for R² confirmed these values’ accuracy. According to this work, computational optimization proves effective in optimizing development and property tailoring for application in industries including aerospace and electrical engineering.

In recent years, rapid advancements in technology, particularly in the realm of artificial intelligence, have significantly transformed the landscape of lung cancer diagnosis. Early detection of lung cancer is pivotal in enhancing patient outcomes; however, traditional diagnostic methods are laborious and time-consuming. Leveraging the power of deep learning techniques, specifically utilizing established neural network architectures, offers a promising solution. This study focuses on the classification of lung images from computed tomography (CT) scans into cancerous and non-cancerous categories. By employing prevalent deep learning models, transfer learning, and rigorous evaluation metrics, this study aims to assess the efficiency of these models in accurately diagnosing lung cancer. The study uses a publicly available dataset and employs preprocessing and segmentation techniques to prepare the images for analysis. The performance of the deep learning models is evaluated on the basis of parameters such as accuracy, sensitivity, specificity, and F1 score. The results demonstrate remarkable accuracy rates, with specific architectures such as ResNet-152V2 and the proposed deep convolutional neural network architecture achieving a staggering 99.1% accuracy. These findings underscore the potential of deep learning techniques in revolutionizing lung cancer diagnosis, offering valuable support to healthcare professionals, and paving the way for more efficient and accurate diagnostic practices.

Manufacturing microscale textures can reduce friction and wear compared with smooth surfaces. In laser surface texturing, the resulting texture geometry is directly affected by laser parameters such as laser power, scanning speed and overlap rate. Therefore, finding the optimal parameter settings to improve the process performance is crucial. One of the biggest challenges encountered during the optimization stage is accurately determining the functional relationship between surface quality and process parameters. Another significant problem is solving the nonlinear equation system. This study aims to investigate the effect of laser-textured surfaces on the tribological behavior of titanium Ti6Al4V. Optimization of laser texture process parameters using metaheuristic algorithms is a hot topic. The first contribution of this study is the modeling of the contact angle, kurtosis and skewness quality parameters of Ti6Al4V titanium alloy based on the polynomial curve fitting method as a function of laser power, laser speed and overlap rate. Another significant contribution is to optimize the proposed nonlinear regression models using metaheuristic approaches such as genetic algorithm, particle swarm optimization, whale optimization algorithm, gray wolf optimization, equilibrium optimizer, teaching-learning-based optimization, student psychology-based optimization (SPBO), slime mould algorithm and social network search. The reason for using many metaheuristics is to test the effectiveness of different metaheuristic methods in solving this problem. Analysis of variance was performed to evaluate experimental data and compare the success of optimization methods. The results revealed that the proposed metaheuristic methods can be applied to the related problem. It has been observed that the methods give identical results. Among these methods, the SPBO method provided the best performance in the optimization of all laser output parameters.

This study fixates on determining the optimum laser input parameters that simultaneously meet the desired kerf width and depth during CO2 laser cutting of various polymethylmethacrylate (PMMA) sheets. It has three contributions. The first is to model the cutting process of PMMA by polynomial curve fitting as a function of laser power, laser speed, and standoff distance. R-squared (R²), adjusted R² and root-mean-square error (RMSE) criteria were taken into account to measure the performance of the proposed model. The effect of laser parameters on the process is investigated by analysis of variance (ANOVA) and sensitivity analysis. The second is to optimize the derived nonlinear regression models using genetic algorithm (GA), particle swarm optimization (PSO), whale optimization algorithm (WOA) and ant lion optimization (ALO) metaheuristic methods and compare the performances of the algorithms. The third is to compare the adequacy of the optimization process with the artificial neural network (ANN). The investigations exhibited that the best-fitting polynomials are obtained with the R² and adjusted R² values of over 99% and 97%, respectively. ANOVA and sensitivity test revealed that the sensitivity of the laser power, which is the most effective parameter, was 150 at low powers and decreased to 0 as the power value increased. When the nozzle distance is 4.1, the proposed metaheuristics gave effective and sufficiently accurate results. PSO stood out in terms of both best cost value (3.49 × 10^–8) and relative error value (0.19%). The relative error of the ANN was found as 3% in terms of kerf depth.

Multilevel inverters (MLI) are popular in high-power applications. MLIs are generally configured to have switches reduced by switching techniques that eliminate low-order harmonics. The selective harmonic elimination (SHE) method, which significantly reduces the number of switching, determines the optimal switching moments to obtain the desired output voltage and eliminates the desired harmonic components. To solve the SHE problem, classical methods are primarily employed. The disadvantages of such methods are the high probability of trapping in locally optimal solutions and their dependence on initial controlling parameters. One solution to overcome this problem is the use of metaheuristic algorithms. In this study, firstly, 22 metaheuristic algorithms with different sources of inspiration were used to solve the SHE problem at different levels of MLIs, and their performances were extensively analyzed. To reveal the method that offers the best solution, these algorithms were first applied to an 11-level MLI circuit, and six methods were determined as a result of the performance analysis. As a result of the evaluation, the outstanding methods were SPBO, BMO, GA, GWO, MFO, and SPSA. As a result of the application of superior methods to 7-, 11-, 15-, and 19-level MLIs according to the IEEE 519—2014 standard, it has been shown that BMO outperforms in 7-level MLI, GA in 11-level MLI, and SPBO in 15- and 19-level MLIs in terms of THD, while in terms of output voltage quality, GA in 7-level MLI, BMO in 11-level MLI, GA and SPSA in 15-level MLI, and SPSA in 19-level MLI come forward.