Language model pre-training architectures have demonstrated to be useful to learn language representations. bidirectional encoder representations from transformers (BERT), a recent deep bidirectional self-attention representation from unlabelled text, has achieved remarkable results in many natural language processing (NLP) tasks with fine-tuning. In this paper, we want to demonstrate the efficiency of BERT for a morphologically rich language, Turkish. Traditionally morphologically difficult languages require dense language pre-processing steps in order to model the data to be suitable for machine learning (ML) algorithms. In particular, tokenization, lemmatization or stemming and feature engineering tasks are needed to obtain an efficient data model to overcome data sparsity or high-dimension problems. In this context, we selected five various Turkish NLP research problems as sentiment analysis, cyberbullying identification, text classification, emotion recognition and spam detection from the literature. We then compared the empirical performance of BERT with the baseline ML algorithms. Finally, we found enhanced results compared to base ML algorithms in the selected NLP problems while eliminating heavy pre-processing tasks.

Identification and location of defects in software projects is an important task to improve software quality and to reduce software test effort estimation cost. In software fault prediction domain, it is known that 20% of the modules will in general contain about 80% of the faults. In order to minimize cost and effort, it is considerably important to identify those most error prone modules precisely and correct them in time. Machine Learning (ML) algorithms are frequently used to locate error prone modules automatically. Furthermore, the performance of the algorithms is closely related to determine the most valuable software metrics. The aim of this research is to develop a Majority Vote based Feature Selection algorithm (MVFS) to identify the most valuable software metrics. The core idea of the method is to identify the most influential software metrics with the collaboration of various feature rankers. To test the efficiency of the proposed method, we used CM1, JM1, KC1, PC1, Eclipse Equinox, Eclipse JDT datasets and J48, NB, K-NN (IBk) ML algorithms. The experiments show that the proposed method is able to find out the most significant software metrics that enhances defect prediction performance.

In 5G vehicular networks, two key challenges have become apparent, including end-to-end delay minimization and data privacy. Learning-based approaches have been used to alleviate these, either by predicting delay or protecting privacy. Traditional approaches train machine learning models on local devices or cloud servers, each with their own trade-offs. While pure-federated learning protects privacy, it sacrifices delay prediction performance. In contrast, centralized training improves delay prediction but violates privacy. Existing studies in the literature overlook the effect of training location on delay prediction and data privacy. To address both issues, we propose a novel genetic algorithm optimized federated learning (GAoFL) approach in which end-to-end delay prediction and data privacy are jointly considered to obtain an optimal solution. For this purpose, we analytically define a novel end-to-end delay formula and data privacy metrics. Accordingly, a novel fitness function is formulated to optimize both the location of training model and data privacy. In conclusion, according to the evaluation results, it can be advocated that the outcomes of the study highlight that training location significantly affects privacy and performance. Moreover, it can be claimed that the proposed GAoFL improves data privacy compared to centralized learning while achieving better delay prediction than other federated methods, offering a valuable solution for 5G vehicular computing.

This paper presents a majority vote decision fusion system called AIVCP (Automated Identification of Vertebral Column Pathologies). With this aim, we proposed a three-step decision fusion algorithm: In the first step, a pool of algorithms from different groups is obtained and the number of classifiers is decreased to 10 with the use of prediction accuracy and classifier diversity concept. As a second step, different majority vote combinations of 10 algorithms are searched with a grid search strategy guided on top of 10-fold cross validation evaluation and with prediction error analysis. In the second step, we obtained four base classifiers, i.e., Naïve Bayes (NB), Simple Logistics (SL), Learning Vector Quantization (LVQ) and Decision Stump (DS) whose majority vote decision fusion generate the most accurate diagnosis rate in Vertebral Column Pathologies domain. As the third step, we applied a Support Vector Machine based feature selection to increase prediction performance of the proposed system further. The experiments are evaluated with the use of 10-fold cross-validation, Sensitivity, Specificity and Confusion Matrices. The experimental results have shown that NB, SL, LVQ, and DS as single classifiers generate 82.58%, 87.09%, 82.90%, and 77.41% average diagnosis accuracies respectively. On the other hand, majority vote decision fusion of these single predictors produces 90.32% accuracy that is higher than each of the constituents. The resultant diagnosis accuracy of Vote algorithm for Vertebral column pathologies is quite promising.

Modern energy grids, with their regional diversity and complex consumption patterns, require accurate short-term forecasting for operational efficiency and reliability. This study introduces a Stacking Ensemble Forecasting (SEF) framework for multi-region household energy demand, utilizing an optimized stacking ensemble model tuned via Bayesian Optimization to achieve superior predictive accuracy. The framework significantly improved accuracy across Diyarbakır, Istanbul, and Odemis, with a final model demonstrating up to 16.47% RMSE reduction compared to the best baseline models. The final model’s real-world performance was validated through a Simulated Digital Twin (SDT) environment, where scenario-based testing demonstrated its robustness against behavioral changes, data quality issues, and device failures. The proposed SEF-SDT framework offers a generalizable solution for managing diverse regions and consumption profiles, contributing to efficient and sustainable energy management.