Artificial intelligence in education (AIED) has expanded rapidly since the emergence of generative artificial intelligence, yet recent development remains insufficiently mapped. This study examines publication growth, thematic change, citation impact, journal profiles and collaboration structure in AI-related research published in four selected journals between 2020 and 2025: Interactive learning environments (ILE), Computers and education: artificial intelligence (CAEAI), Journal of learning analytics (JLA) and International journal of artificial intelligence in education (IJAIED). Because three venues also publish broader educational technology research, 2719 Scopus records were screened using an explicit AI-relevance criterion; 1397 publications were retained and 1322 excluded. AI-relevant output rose from 54 publications in 2020 to 558 in 2025, a 10.3-fold increase and 59.5
Education typically brings together three primary stakeholders: the government, teachers, and students. These actors collectively form the foundation of a country’s educational system, each contributing in their way to the enhancement of education quality. However, student voices are often underrepresented in decisions regarding teaching standards, a situation particularly evident in the academic ecosystem of the Democratic Republic of Congo (DRC). One major gap in the DRC’s educational landscape is the lack of a comprehensive report reflecting computer science students’ feedback on key aspects of teaching quality. This study seeks to fill that gap by leveraging Natural Language Processing (NLP) techniques to analyze student perceptions of computer science education, using responses to open-ended survey questions focused on infrastructure and teaching quality, faculty, and curriculum. Sentiment analysis was applied to gauge the emotional tone of responses from students across 18 leading universities and higher educational institutions in the DRC, revealing a plurality of positive sentiment, though accompanied by substantial neutral responses (29.3
Ultra-high voltage (UHV) transmission systems require shielding clothing with high conductivity, uniform structure, and stable electromagnetic protection to ensure operator safety. Microscopic defects like pinholes, micro-cracks, fiber fractures, and coating discontinuities severely reduce effectiveness and are difficult to detect with conventional manual inspection. This suggested model proves to be an efficient, energy-aware, reliable, and scalable solution for the intelligent quality assessment of the UHV shielding clothing manufacturing and maintenance processes. In particular, the AI-aided Intelligent Detection Model is created to automatize and improve the accuracy of classification of microscopic fabric defects, which can be used to detect fabric defects of UHV shielding clothing materials. For this experiment, the Kaggle Multi-Class Fabric Defect Detection Dataset has been used as a source, it includes 3077 images of microscopic fabric defects belonging to several classes of defects. These images belong to generic images of fabric defects and do not contain any information about UHV shielding clothing, conductive threads, or EM shielding deterioration. In this regard, the following results have been achieved with the accuracy (98.42
Federated Learning (FL) allows for collaborative model training across decentralized clients while maintaining data privacy; yet real-world deployments always include heterogeneous and non-identically distributed (non-IID) client data. Existing FL research primarily assesses system performance using aggregate metrics like global or average accuracy, which can hide differences in client behavior. This study provides a thorough empirical evaluation of federated learning under controlled data heterogeneity, with a major emphasis on worst-client performance as a measure of client-level robustness. Experiments are carried out on MNIST, CIFAR-10, and CIFAR-100 utilizing normal FedAvg training and commonly accepted baselines, such as FedProx and q-FedAvg, at various levels of Dirichlet-based label skew. The results show that, as heterogeneity increases, worst-client accuracy falls quickly, revealing silent errors that were previously disguised by aggregate evaluation. A comparison of mitigation options reveals that server-side client re-weighting improves worst-case performance under modest heterogeneity but becomes unstable and damaging under strong non-IID situations. In contrast, simple client-side data balancing reliably improves worst-client accuracy across all heterogeneity regimes without requiring additional infrastructure or algorithmic complexity. These findings emphasize the limitations of aggregate evaluation in federated learning, as well as the need of client-aware metrics and mitigation measures that increase local update quality while dealing with client-level inequalities in the face of true data heterogeneity.
The rapid migration of electrical grids towards Smart Grid (SG) architectures produces high-dimensional data streams that must be processed intelligently in real time. Traditional modelling, optimisation, and control approaches cannot handle the growing complexity of contemporary power systems. This comprehensive study aims to provide a synopsis of AI applications in smart grid settings for load forecasting, transient and voltage stability analysis, fault detection, and cybersecurity. This work provides a unified analytical framework that links supervised, unsupervised, reinforcement, and ensemble learning techniques to specific smart grid operational problems, in contrast to prior surveys that address these areas separately. The main contributions are (i) a structured AI-smart grid functional taxonomy; (ii) critical analysis of the challenges of data quality, model interpretability, computational scalability and adversarial security; and (iii) a roadmap for explainable, reliable and carbon-neutral intelligent energy systems. Compared with more conventional methods, AI-based solutions often outperform them in prediction accuracy, stability, categorisation speed, and defect identification accuracy. However, there are major challenges in real-world deployment, standardisation and regulatory compliance.
