The purpose of the study was to develop a mechanism for training physical education and sports professionals. The methodology included an analysis of the use of innovative methods such as module-rating training implemented at the Kazakh Academy of Sport and Tourism, programmes for the development of digital competencies of coaches through the Open Sport Academy platform, and dual programmes such as the cooperation of the Vasyl Levsky National Sports Academy with Bulgarian sports federations through the Dual Education for Sports Professionals programme. The study found that the training of physical education and sports professionals should be closely linked to labour market requirements, in particular through the integration of modern technologies and increased practical training. The study of the labour market in countries such as Kazakhstan and Bulgaria showed that successful educational programmes should take into account not only theoretical aspects, but also the opportunity for students to gain real-world experience in sports clubs, federations and other organisations. In addition, it has been found that the introduction of dual educational programmes that combine university studies and practical workplace activities has a significant impact on the competitiveness of graduates. An important aspect is also the use of digital technologies, such as online learning platforms and virtual simulators, which allow students to acquire the necessary skills to work in the context of the digital transformation of the sports industry. The practical significance of the study lies in the fact that the findings can serve as a basis for improving curricula and enhancing cooperation between educational institutions and employers in the field of physical education and sports. 

This study utilizes the Digital Competence Framework for Educators (DigCompEdu) and machine learning (ML) techniques to evaluate and predict the ICT proficiency levels of public school teachers in Bukidnon. Analyzing a dataset of 1,275 responses and addressing data imbalances, several classification models were evaluated to identify the most reliable predictor of teacher competence. The findings indicate that the majority of teachers currently operate at the 'Integrator' (B1) level. Key predictors of proficiency include skills in online safety, collaborative learning, and the creative use of digital tools. Among the tested algorithms, Random Forest emerged as the most effective model for accurately classifying teacher skill levels. This research provides a data-driven roadmap for educational policymakers, offering actionable insights for designing targeted professional development programs that foster transformative teaching and improved student outcomes.

Recently, the academic recommendation system represents the process of suggesting suitable institutions, courses, or learning pathways for students based on their performances and interests. Yet, the conventional systems didn’t concentrate on temporal dynamic pattern analysis within the Indian higher education institutions, leading to less effective or static academic recommendations. Thus, an academic recommendation system is proposed for Indian higher education institutions using Few-Shot PairNorm-Apical Dendrite Graph Attention Networks (FSPN-ADGAT) by considering temporal dynamic pattern analysis. Primarily, the student data undergoes pre-processing. Further, student performance analysis is done, followed by feature extraction. Now, the institutional course data undergoes pre-processing, followed by contextual embedding of text using Adapter Layers-Bidirectional Encoder Representations from Transformers (AL-BERT). Similarly, by using SRC, course similarity is analyzed between the pre-processed course data and extracted features. Similarly, the temporal dynamic pattern analysis is done from the pre-processed course data using Student-t Likelihood-based Bayesian Change Point (SL-BCP) and indicator extraction. Now, based on the analyzed course similarity, extracted features, contextual embedding output, analyzed temporal dynamic patterns, and extracted indicators, the node and matrix construction is performed. Lastly, the academic recommendation using FSPN-ADGAT provides personalized course suggestions to the students. Therefore, the proposed FSPN-ADGAT attained a lower Mean Absolute Error (MAE) of 0.171 than the conventional techniques.

This paper presents an information system developed to automate the systems analysis of the quality of technical students’ training using correlation and regression methods. The article considers key problems of quality assessment and outlines the theoretical foundations of correlation and regression analysis in the context of educational data. The structure and algorithm of an information system designed for automated analysis of educational datasets are presented. The system allows to determine pairs of courses for which prediction of grades by means of regression analysis is performed with minimal error. In this study, grades from courses for the previous period were considered as known parameters x, and grades from courses for the next period were considered as predicted results y. The correlation analysis of educational data involved calculating the Pearson correlation coefficient Corr, which quantitatively describes the linear relationship between two parameters, x and y, in the educational dataset. The correlation coefficient Corr allows for a targeted investigation of relationships with high Corr values. The regression analysis of the data involved constructing a regression equation approximated by a polynomial of degree p to establish the relationship between the x and y parameters of the educational dataset. The accuracy of the approximation was evaluated using the root mean square error (Rmse) for the training set and RmseV for the validation set. The automatic selection of the polynomial degree pA, was performed according to the criterion of minimizing the approximation error RmseV on the validation dataset, while also ensuring the monotonicity of the regression equation. Developed in Python, the software performs correlation and regression analysis, prediction, outlier detection, and result visualization. This approach was applied to analyze the semester grades of students in the 'Computer Science' program, covering 12 courses over the first four semesters. Using the constructed regression equations, were forecasted students’ grades in six courses for the 3rd and 4th semesters based on their performance in the same courses during the 1st and 2nd semesters. The developed regression model also allows for evaluating students’ academic achievements through the outlier detection. The proposed correlation and regression analysis models are highly scalable, enabling the processing of educational data for large size. Integrating correlation and regression methods into the systems analysis of technical education quality allows for automated analysis of educational monitoring data, forecasting of student performance, outlier detection, and the recommendation of elective courses to optimize students’ educational trajectories.

This study examines the integration of computational deep learning and digital literacy from 2011 to June 2025. Employing a hybrid methodology of Systematic Literature Network Analysis and Latent Dirichlet Allocation topic modeling, 141 high impact documents were synthesized following the PRISMA 2020 protocol. Findings reveal a conceptual shift from technical exploration (2011–2018) toward human-centric, pedagogical deep learning frameworks (2019–2025). While publications peaked in 2024, Australia and South Korea emerged as leading centers of excellence in citation impact. Latent Dirichlet Allocation modeling identified ten topics, uncovering a significant research gap in using AI for fundamental research processes compared to its dominance in instructional assessment. This study provides a novel mapping of thematic evolution and offers strategic recommendations for longitudinal empirical studies and inclusive AI-driven pedagogical designs.

This study examines AI-related technological proficiency among undergraduate students at the University of Casablanca and identifies the most informative indicators for prediction. Using a validated 30-item instrument covering AI applications, AI-related skills, and improvement strategies, data were collected from 600 students drawn from science and humanities programs. Overall proficiency was moderate: 63.3% of respondents met the predefined threshold, and significant group differences were observed by gender and academic specialization. For predictive modeling, correlation-based feature selection retained 17 high-value items. Two classifiers were then trained and evaluated using a 75/25 hold-out split, complemented by repeated stratified 10-fold cross-validation to assess stability. The Support Vector Classifier achieved 96.7% test accuracy with AUROC = 0.9666, while Gaussian Naïve Bayes reached 94.7% accuracy with AUROC = 0.9560; cross-validated estimates remained consistent with these results, supporting robustness. These findings indicate that a reduced set of questionnaire items can provide reliable estimates of students’ AI-related technological proficiency and can support scalable assessment and targeted interventions in higher education.

This study investigates the use of AI-driven macro expression analysis to enhance the engagement of hyperactive students in English language learning. By utilizing Convolutional Neural Networks (CNN) and K-Nearest Neighbors (K-NN), this research aims to detect and analyze students' macro facial expressions, as well as their correlation with engagement levels. Data was obtained from 24 learning videos, consisting of 13,263 frames, analyzed to identify expressions of boredom, sadness, and happiness. The analysis results show that boredom and sadness dominate, while happiness is recorded at a lower frequency, indicating the need for a more varied and responsive teaching approach. This study also finds that AI-driven emotion detection can provide more adaptive feedback for hyperactive students, allowing teachers to adjust teaching methods in real-time according to the students' emotional responses. The findings contribute new insights into the field of inclusive education by integrating AI technology to monitor and tailor learning for students with special needs. Theoretically, this research enriches the understanding of the role of macro expressions in student engagement, particularly in the context of ADHD. Practically, the results offer technology-based solutions to support more adaptive and responsive teaching that aligns with students' emotional changes. This research contributes to the development of more holistic and interactive learning methods, which can improve learning outcomes for students with special needs, especially in English language education.

Curriculum mapping for AUN-QA is often time-consuming and prone to inconsistency because learning-outcome evidence is scattered across multiple courses and program documents. This study examines the reliability of large language models in supporting course learning outcome–program learning outcome (CLO–PLO) alignment decisions for AUN-QA–aligned curriculum mapping. Using Mechatronics Engineering curriculum materials (2022–2024), AUN-QA v4.0 indicators were operationalized into a structured evidence schema to guide prompting and interpretation. GPT-4 produced 120 CLO–PLO alignment decisions, which were independently annotated by two domain experts and adjudicated by a third expert to establish reference labels. Model–reference agreement was evaluated using precision, recall, F1-score, and Cohen’s kappa (κ), yielding 0.89, 0.85, 0.87, and 0.81, respectively. We also developed a dashboard that summarizes alignment coverage and supports PDCA-based curriculum improvement by flagging potential gaps and redundancies. The findings suggest that LLM-assisted alignment can reduce mapping workload and improve auditability while remaining consistent with expert judgment, enabling more scalable evidence-based AUN-QA evaluation.

Identifying students’ learning styles is an important step in supporting adaptive and personalized learning environments, particularly in higher education contexts. This study investigates the use of a transformer-based language model, DistilBERT, for automated learning style identification based on the Felder–Silverman Learning Style Model (FSLSM). Binary responses from the 44-item Index of Learning Styles (ILS) questionnaire were systematically transformed into descriptive textual representations, preserving the underlying FSLSM decision logic while enabling semantic modeling. Separate DistilBERT classifiers were fine-tuned for the four FSLSM dimensions—Active/Reflective, Sensing/Intuitive, Visual/Verbal, and Sequential/Global. Model training was conducted using three epochs within a stratified five-fold cross-validation framework, with the primary objective of assessing optimization stability and representational consistency rather than predictive superiority. Across all dimensions, training loss decreased monotonically, with the most substantial reductions occurring between the first and second epochs, indicating rapid adaptation of pre-trained representations to the structured learning style descriptions. Differences in convergence behavior across dimensions were observed, reflecting variation in class distributions and response patterns. Near-perfect classification metrics were obtained across multiple folds; however, these results are interpreted as evidence of deterministic consistency between the textual inputs and FSLSM-derived labels rather than independent generalization performance. To examine alignment with human judgment, model predictions were further compared with expert assessments conducted independently using standard FSLSM interpretation guidelines. The resulting Cohen’s Kappa coefficient of 1.0 indicates perfect agreement, confirming that the model faithfully reproduces expert-consistent FSLSM categorizations under controlled conditions. Overall, the findings demonstrate that transformer-based models can reliably encode and recover rule-based learning style constructs from descriptive questionnaire data, supporting their use as computational tools for scalable, consistent learning style analysis, while acknowledging the task's deterministic and framework-dependent nature.

As Large Language Models (LLMs) become increasingly embedded in intelligent tutoring systems (ITSs), a growing need exists to understand how learners engage with these tools, especially in cognitively demanding domains such as computer programming. While prior research has focused mainly on LLM-generated scaffolding, less attention has been paid to student-side engagement, including how learners think, respond, and regulate their learning during natural language tutoring sessions. This study addresses that gap by identifying learner engagement state profiles based on behavioral and metacognitive patterns in student dialogue. Drawing on data from 36 recorded LLM-mediated C introductory programming tutorials, 1,046 dual-annotated student utterances were analyzed using K-means clustering. The analysis revealed three distinct learner engagement state profiles: Passive Reactors, Clarification Seekers, and Reflective Performers. These engagement states differed in response accuracy, metacognitive expression, and interaction style. Passive Reactors showed low initiative and limited self-regulation; Clarification Seekers demonstrated moderate accuracy and reactive help-seeking; Reflective Performers exhibited strategic engagement and high metacognitive activity. This study introduces an exploratory, scalable approach to learner profiling through natural language dialogue, advancing the design of adaptive, learner-aware LLM tutoring systems. The findings support the development of real-time learner modeling techniques that move beyond correctness, offering actionable insights for delivering more personalized and effective AI-assisted instruction in programming education.

The efficient allocation of finite resources to a dynamic patron base represents a core challenge in modern library management. Traditional heuristic approaches often lack the formal rigor needed for verifiable optimization and proactive planning. This paper introduces a novel formal framework grounded in automata theory to model library operations, patron behavior, and resource allocation strategies. We define a Library Resource Automaton (LRA), a deterministic finite automaton whose states represent distinct configurations of resource availability, whose input alphabet encapsulates patron interactions, and whose transition function formally encodes allocation policies. By interpreting sequences of patron actions as strings in a formal language, the LRA provides a computationally tractable and analytically powerful model for simulating library states, predicting bottlenecks, and synthesizing optimal allocation strategies. We elaborate on the theoretical foundations of the model, present a detailed multi-layer automata architecture for handling complex, multi-resource scenarios, and discuss algorithms for state space analysis and policy optimization. Furthermore, we explore the integration of temporal logic for specifying and verifying critical system properties such as fairness and liveness. This work establishes a rigorous bridge between theoretical computer science and library information science, offering a new paradigm for building predictable, efficient, and patron-centric library management systems.

An accurate and comprehensive assessment of student engagement in classrooms is crucial for enabling data-driven teaching and personalized education. Current approaches primarily rely on teacher observation or student self-reports, which are often subjective, delayed, and unable to capture cognitive engagement. To address these limitations, this study proposes a Multimodal Cognitive-Attention Fusion (MCA Fusion) framework, grounded in Fredricks’ three-dimensional engagement model.  The framework integrates electroencephalography (EEG), facial expressions, and body posture to simultaneously quantify cognitive, emotional, and behavioral engagement.  Built on a Transformer architecture, it employs self-attention to extract temporal features within each modality and introduces a cognition-guided cross-attention mechanism to dynamically integrate multimodal signals. To validate the framework, experiments were conducted with 36 undergraduate students in real classroom settings. The results demonstrate that our framework significantly outperforms all single-modality baselines, achieving an accuracy of 92% and an F1-score of 94.87%. Compared with the best single-modality model (EEG), the F1-score improves by 34.58 percentage points. Ablation studies further confirm the critical role of the cognitive modality (EEG) and the MCA Fusion mechanism, the removal of which leads to F1-score reductions of 62.58 and 56.16 percentage points, respectively. The proposed approach not only provides a theoretically informed and technically evaluated framework for engagement recognition but also provides a methodological foundation for future closed-loop “perception–assessment–feedback” systems in intelligent learning environments.