This study analyzes the artificial intelligence (AI) literacy levels of music education students and explores how future music educators perceive and integrate AI tools within creative and instructional contexts. The research group consisted of 143 students enrolled in music education programs from different faculties, such as Education, Fine Arts, and Conservatory. Participants were selected through a purposeful sampling technique, ensuring inclusion of individuals likely to encounter AI in educational and performance settings. The study employed a causal-comparative (ex post facto) design, which enables examination of naturally existing groups differing in characteristics such as prior exposure to or familiarity with AI technologies. Quantitative data were analyzed to compare students’ AI literacy levels across demographic variables, including gender, type of high school graduated from, academic achievement level, and prior knowledge of AI. However, it was determined that the artificial intelligence literacy level of music teacher candidates who have knowledge about artificial intelligence programmes and use artificial intelligence programmes in music education is significant. Results indicated that students’ AI literacy levels were at a medium level. No significant differences were found in AI literacy by gender, type of high school graduated from, academic achievement, or frequency of internet use. However, significant differences were observed among students from different faculties, as well as among those who had prior knowledge of AI programs, used AI applications in music and music education, and used them more frequently. These findings suggest that direct engagement with AI tools contributes to higher literacy levels.

Identifying suicidal ideation on Twitter is crucial for timely intervention and suicide prevention efforts. This study leverages Twitter data to identify individuals contemplating suicide, addressing the challenge of distinguishing genuine suicidal ideations from those posted for entertainment. Given the rise in depression and anxiety, particularly among the youth, an ensemble model combining various machine learning algorithms is proposed to enhance the accuracy of detecting genuine suicidal tweets. The ensemble model integrates the predictions of multiple internal models: Logistic Regression, LinearSVC, XGBoost, Naive Bayes, and Random Forest. Each model generates a prediction based on extracted features from the tweets, and the final prediction is determined by calculating the weighted average of these predictions, considering the relative importance of each model. If this weighted average exceeds a predefined threshold, the tweet is classified as non-suicidal; otherwise, it is classified as suicidal. This methodology allows the ensemble model to balance the strengths and weaknesses of individual models, resulting in a robust and accurate classifier. Incorporating BERT classification and VADER sentiment analysis, the model is trained on labeled data to capture intricate patterns in tweet embeddings. Evaluated against various performance metrics, the ensemble model achieves an accuracy of 95.81%, precision of 93.45%, recall of 89.90%, and an F1-score of 94.68%, significantly outperforming individual models. The model also demonstrates a superior AUC-ROC value of 0.95, indicating excellent performance in distinguishing between classes. This approach not only advances the current methodologies but also contributes to public health by enhancing the reliability of suicide prevention efforts on social media platforms. 

We present SECURE-XED, a unified, explainable, and adversarial-aware learning system for two traditionally disjoint domains: Android malware classification and deepfake detection. The system uses a shared convolutional neural network backbone with domain-specific input adaptation and two task-specific heads. It integrates SHapley Additive exPlanations for explanation, the Fast Gradient Sign Method and Projected Gradient Descent for adversarial evaluation under Structural Similarity Index Measure guardrails, and a selective-activation controller that enables behavioral malware features when confidence is low or evasion is suspected. A hybrid Seagull Optimization Algorithm–Imperialist Competitive Algorithm procedure jointly tunes clean accuracy, robustness under projected-gradient attacks, expected calibration error, parameter count, and inference latency. SECURE-XED is evaluated on standard Android malware corpora and the FaceForensics++, Celeb-DF, and Deepfake Detection Challenge benchmarks. Under Projected Gradient Descent, Android malware accuracy decreases by approximately 19–21% with a perturbation budget of 0.1 over 40 steps, while deepfake accuracy decreases by 16.6–17.6% with a perturbation budget of 0.03 over 10 steps, demonstrating domain-dependent adversarial sensitivity. Under clean deepfake inference conditions, excluding post-prediction explanation generation and adversarial processing, the unified configuration reduces average inference latency from 34.1 to 26.9 milliseconds per frame and approximately halves the parameter requirement relative to maintaining separate model instances, while retaining equal or slightly better clean accuracy. A classroom-oriented simulator exposes explanation overlays and adversarial controls on real frames. A six-participant human-grounded pilot descriptively showed a 22-percentage-point change in Region Identification Accuracy, a 12-second reduction in decision time, and a 0.9-point Likert change in perceived clarity and trust; these pilot outcomes are descriptive and not inferential. The resulting framework combines shared cross-domain representation, multi-objective optimization, adversarial evaluation, explainability, and model-in-the-loop educational interaction within a single modular architecture.

Aspect-Based Sentiment Analysis (ABSA) has become an integral component of Natural Language Processing (NLP). It provides comprehensive insights into individuals' sentiments regarding specific aspects. This study conducts a comprehensive bibliometric analysis of 913 journal articles published from 2010 to 2024, sourced from the Scopus database, to examine trends, challenges, and prospective directions in ABSA research. The research examines the expansion of publications, citation metrics, and scholarly networks. It integrates performance analysis metrics (such as h-index, g-index, and citations per paper) with sophisticated science mapping techniques, including keyword co-occurrence networks, thematic evolution, and thematic mapping. This novel integrated approach, rarely employed in prior ABSA bibliometric studies, reveals both historical trends and emerging niche themes that remain underexplored. The findings illustrate the evolution of ABSA from rule-based methodologies to transformer-based architectures applicable in e-commerce, social media, and customer feedback systems. Key issues identified include multilingual adaptability, implicit sentiment detection, and cross-domain scalability. The research indicates that global institutions have significantly contributed, yet productivity and impact differ markedly across nations. Multimodal analysis, transfer learning, and contextualized models such as BERT are pivotal contemporary methodologies that can assist in addressing existing challenges. Future research should concentrate on integrating diverse disciplines, analyzing data across various languages and modalities, and developing scalable and comprehensible models. This study contributes to the domains of artificial intelligence and sentiment analysis by offering a comprehensive and data-driven overview of the ABSA landscape. It accomplishes this by providing both strategic insights and methodological enhancements.

This article presents a set of specialized software tools for educational data mining and describes the integration of the developed tools into the educational process of students of technical specialties. The developed set consists of three programs that implement the main tasks of educational data mining. Program No. 1 provides segmentation, analysis, and visualization of images of educational materials; Program No. 2 performs data clustering; Program No. 3 performs correlation and regression analysis to identify relationships and predict learning outcomes. The developed programs can be used to implement a systematic analysis of education quality, during which automated processing of educational data and the formation of recommendations for improving the educational process are performed. The programs are implemented in Python. A scheme for integrating specialized programs into the educational process has been developed. The scheme reflects the relationships between the programs that process educational materials, analyze the results of the educational process, influence students’ educational trajectories and the educational process.
Program No. 1, “SegmentFuzzy24,” is designed for segmentation, analysis, and visualization of images of educational materials. Such processing makes it possible to highlight the investigated elements of technical devices in order to focus students’ attention, as well as to determine the quantitative and geometric characteristics of multimedia presentation slides, including the number and size of symbols, the area of graphical objects, etc. Segmentation is performed using the region growing method. The determined characteristics are used for the classification of educational materials or are transferred to Programs No. 2 and No. 3 for further cluster, correlation, and regression analysis.
Program No. 2, “ClusterFuzzy23,” is designed for clustering educational data, including students’ learning outcomes and the characteristics of educational materials obtained from Program No. 1. Objects are divided into groups with similar characteristics using the K-means method. During the clustering process, the sum of the squared Euclidean distances between each object and the centroid of its cluster is minimized. Data clustering ensures their differentiated processing and contributes to the adjustment of educational trajectories. The clustering results can be transferred to Program No. 3 for further analysis. Fuzzy membership functions are used to determine the degree of membership of objects in overlapping clusters.
Program No. 3, “CorrelRegres25,” is designed for correlation and regression analysis of educational data obtained directly or as a result of clustering in Program No. 2. Correlation analysis is used to establish relationships between the characteristics of the educational process, while regression analysis is applied to model their dependencies using polynomial functions and to predict learning outcomes. The obtained models are used to adjust educational trajectories, while objects that have significant deviations from regression dependencies are classified as outliers.
The developed programs have been demonstrated on sample data to process real educational data, namely for image segmentation and visualization, clustering of educational data, and their correlation and regression analysis.
Data clustering and correlation–regression analysis were performed using the grades of N0 = 76 students in 12 subjects, assessed on a 100-point scale, for courses completed during the first and second years of study. The initial sample of N0 students was randomly divided into a training set (N = 60, 80% of the data) and a validation set (NV = 16, 20% of the data). The quality of the clustering results was evaluated using the Silhouette coefficient, Calinski–Harabasz index and Davies–Bouldin index. The analysis indicated that three or four clusters provide an appropriate representation of the data. The optimal polynomial degree pA = 3 for the regression model was determined based on the minimum root mean square error RmseV = 12.37458, obtained on the validation set. The relatively high RmseV value can be attributed to the influence of a considerable number of factors affecting students’ academic performance. The resulting regression model was used to predict students’ grades in a subject for the subsequent academic period based on their grades in the same subject during the previous period. A promising direction for further improvement of the developed prediction system is the integration of artificial neural networks.

Teacher education requires models that connect reflective practice and digital competence with the psychological processes through which professional growth occurs. This longitudinal pre-test/post-test quasi-experiment examined whether a composite reflective-digital practice process predicted professional development and whether instructional smart self-efficacy mediated this association. The sample comprised 120 third-year pre-service teachers enrolled in a Computer Science Pedagogy program (N = 120) who completed a 15-week intervention organized across four instructional conditions. Data were collected with an expert-reviewed 19-item questionnaire using a five-point Likert scale. Confirmatory factor analysis and structural equation modeling were used to assess the measurement and structural models, and indirect effects were evaluated with 5,000 bias-corrected bootstrap resamples. Internal consistency was satisfactory across the four constructs (Cronbach’s alpha = 0.86-0.91). Reflective-digital practice was positively associated with professional outcomes (beta = 0.41, p < 0.001), whereas its path to instructional smart self-efficacy (beta = 0.12, p = 0.184) and the direct path from self-efficacy to professional outcomes (beta = 0.10, p = 0.211) were not significant. The bootstrapped indirect estimate was significant (B = 0.09, 95% CI [0.02, 0.17], p = 0.018), and the model explained 54% of the variance in professional outcomes (R² = 0.54). These findings provide preliminary empirical support for a mediated reflective-digital pathway, while the single-program sample and reliance on self-reports limit causal and population-level inference.

Sentiment analysis of medication evaluations uses Natural Language Processing (NLP) techniques to analyze users' thoughts and feelings about medications (treatments) they have received, helping evaluate the effectiveness of therapy. Machine learning aids in evaluating medication effectiveness; however, it has limitations due to the context-dependent nature of medication-related sentiment. Even though deep learning provides a better ability to capture nuances in language, deep learning models will require larger datasets and considerable computational resources. A Hybrid Bidirectional Gated Recurrent Unit-Convolutional Neural Network (Bi-GRU-CNN) is proposed for sentiment analysis of medication evaluations to address these shortcomings. The evaluation of medicines from patients involved collecting reviews and compiling them into a structured dataset as the first step. The process uses various methods to preprocess the reviews, including lowercasing, stop-word removal, and punctuation removal. SentiWordNet is used to assign sentiment polarity to reviews after they have been pre-processed, classifying them as neutral, negative, or positive. The attention-based Bidirectional Long Short-Term Memory (Bi-LSTM) model is then applied for part-of-speech tagging to extract semantic information, and a dependency tree parser that uses Global Vectors for Word Representation (GloVe) to extract syntactic information. Bidirectional Encoder Representations from Transformers (BERT) are then used to fuse the two features into a single representation. A hybrid BiGRU-CNN model is then used to classify sentiment and accurately predict the polarity of drug reviews. The CNN layer captures many local patterns, and the BiGRU layer processes sequential dependencies in both directions. The features are combined and utilized in a series of dense layers to create classifications. The proposed method achieved an overall Accuracy of 97.50%, a False Omission Rate of 4.30%, and a selectivity of 94.20% in sentiment analysis of drug reviews. The proposed method improves contextual understanding and feature extraction for a more granular understanding of sentiment.

The article presents a comprehensive theoretical and empirical analysis of the research competence of a modern teacher as an integral component of professional activity and a key factor in improving the quality of education. Based on the generalization of scientific approaches and the results of modern psychological and pedagogical research, the content of the concept of “research competence” is clarified, its five-component structure is determined, and the role of the psychological characteristics of the teacher – temperament type, internal motivation, reflexivity – is substantiated in forming readiness for research activities. An empirical study conducted among 10 teachers of general secondary education institutions made it possible to quantitatively assess the level of formation of professional competencies. It was found that the highest average indicators are observed in communicative and methodological competencies, while digital, inclusive, managerial and especially research competencies require significant development. Analysis of individual psychological characteristics showed that 54% of respondents have an extroverted type of temperament, which correlates with higher communicative and organizational indicators, while 46% of introverts demonstrate higher analytical and methodological abilities. Correlation analysis confirmed the presence of statistically significant relationships between research competence and internal professional motivation, as well as its moderate relationship with methodological competence. Economic and mathematical modeling allowed us to formalize the level of research competence as an integral indicator and assess its impact on the effectiveness of pedagogical activity. It was found that an increase in the integral index of research competence by 0.1 conventional unit causes an increase in pedagogical results by an average of 8-12%, which confirms the importance of research activity for improving the quality of education. Based on the data obtained, an author’s three-component model of the development of research competence is proposed, which involves the integration of research practices into the content of pedagogical education, the creation of school laboratories and professional communities, and the formation of a system of motivational incentives. It is concluded that the development of research competence is a strategic condition for the formation of a culture of evidence-based learning, ensuring innovative development of schools, and improving the quality of education in Ukraine.

Predicting student academic performance accurately is essential and allows for timely interventions and data-supported educational decision-making. However, the black-box nature of state-of-the-art machine learning models hinders their use in policy-sensitive educational settings, where interpretability and accountability are paramount. We propose a structured hybrid explainable artificial intelligence (XAI) framework that combines three tree-based ensemble models–namely, random forest, XGBoost, and CatBoost–using a prediction-level soft-voting strategy and couples them with complementary post-hoc interpretation methods at the explanation level (SHAP and LIME). Instead of proposing a new base algorithm, this study systematically unifies heterogeneous boosting and bagging approaches with strict validation as well as dual-layer explanation consistency evaluation. Experiments on the UCI Student Performance dataset show that the proposed framework can achieve a competitive predictive performance (91.8% accuracy, ROC–AUC = 0.953) together with transparent and actionable interpretability. The robustness of the interpretability layer is further supported by a quantitative assessment of explanation stability, fidelity, and agreement across various methods. The presented framework harmonizes the trade-off between accuracy and interpretability to provide a deployable, policy-aware decision-support solution that is aligned with responsible AI principles for adoption within educational contexts.

The accurate energy consumption forecast of educational institutions is critical for sustainable energy management and intelligent campus infrastructure. Classical forecasting models fail to simultaneously capture local temporal variations and long-range dependencies, thus diminishing their reliability across different time horizons. To overcome this, we propose an effective hybrid deep learning method that combines Convolutional Neural Networks (CNNs) with Long Short-Term Memory (LSTM) networks and domain-driven feature engineering to incorporate occupancy and seasonality patterns. The dataset from an office building (research support facility) was used to perform a series of short-term (05-minutes), medium-term (daily), and long-term (weekly) forecasts. Experimental results show that the proposed forecasting approach demonstrates greater precision than the ten baselines including machine learning and deep learning methods in short, medium, and long time horizons. In particular, the model with feature engineering improved Root Mean Squared Error (RMSE) by 4.71% and Mean Absolute Percentage Error (MAPE) by 3.99% in the short term and saw even greater improvements for medium-term forecasts, with RMSE improved by 6.08% and MAPE by 5.37%, and for long-term forecasts, with RMSE improved by 2.58% and MAPE by 0.33%, respectively. The outcomes here offer significant benefits to the science community, supporting the development of such smart campuses with a specific, understandable methodology for energy forecasting. The methodology enables the application of data-driven energy management, cost optimization, and sustainability planning in educational spaces. The findings also provide valuable insights for deploying Artificial Intelligence (AI)-driven infrastructure management systems in academia and institutional energy governance.

Public digital services in developing economies have become central to citizens’ access to government functions, yet frequent deployment failures undermine trust and service delivery. This study investigated whether multilingual citizen feedback could serve as an early-warning signal for deployment failures in resource-constrained public sector digital services. A bimodal failure detection framework was developed combining English system logs with citizen feedback in Akan, Ewe and Ga languages from a national citizen service portal in Ghana handling over 50,000 monthly users. Rather than building language-specific classifiers requiring annotated training data, the study leveraged instruction-tuned large language models as general-purpose interpreters using few-shot prompting. Analysis of 18 months of deployment logs and 1,217 citizen feedback messages revealed that a fusion model combining XGBoost for log analysis and prompt-engineered Qwen 2.5-3B-Instruct for feedback interpretation achieved 82 per cent recall and 76 per cent precision, representing a 21 percentage point improvement over log-only approaches. Citizen feedback preceded technical alerts in 32 per cent of failure cases, with an average detection advantage of 2.3 hours, which increased to 3.8 hours during peak usage periods. This study provides empirical evidence that instruction-tuned large language models can extract operationally useful failure signals from low-resource languages and improve post-deployment early warning when fused with conventional log-based monitoring. The findings imply that public institutions in multilingual developing contexts can adopt LLM-based monitoring without requiring annotated datasets or language-specific processing pipelines.

The article presents the results of a 2023 study carried out at the Institute of Social Sciences with the aim of assessing the behavioral characteristics, habitual communication practices and personal adaptation experience of medical students in the virtual space. The study was conducted using the method of narrative interviews, in which 50 students in their 4th–6th years of study at Sechenov University took part. The study showed that the structure of online activity of medical students reflects a shift in their interests towards educational content, queries and topics related to professional and personal self-development. What the youth audience has in common is the consumption of entertainment content in their free time. The obtained data can be used in the formation of an appropriate educational system for medical students in the virtual space. This certainly expands their ability to receive up-to-date information through social networks.