Modeling the temporal dynamics of student learning is a central goal in educational data mining. Deep Knowledge Tracing (DKT) has emerged as a key approach, yet existing models are highly sensitive to out-of-distribution (OOD) inputs, such as those arising from curriculum changes, new assessment formats, or behavioral noise, which severely degrade predictive reliability. To address this challenge, we propose Energy-Based Out-of-Distribution Deep Knowledge Tracing (EB-OOD DKT), a unified framework that integrates energy-based uncertainty estimation and contrastive representation learning within a transformer-based DKT architecture. The model computes energy scores via the negative log-sum-exponential of prediction logits, serving as confidence indicators for detecting OOD inputs during inference. Additionally, an InfoNCE-based contrastive loss enhances representation robustness by aligning in-distribution samples and separating OOD cases in latent space. Temporal and behavioral context features, such as normalized response intervals and cumulative attempt counts, are incorporated to enrich cognitive-behavioral modeling. Experiments on four public educational datasets demonstrate consistent improvements in prediction accuracy and OOD detection. EB-OOD DKT provides a promising approach for more reliable student modeling across educational platforms with different content distributions.

The purpose of the study. The digitalization of education implies large-scale transformations encompassing the implementation of digital technologies at every level of general and vocational education, as well as additional education, as well as changes in the interactions of all participants in the educational process. One of the six key challenges addressed by the Education Development Strategy to 2036 is the rapid spread of digital technologies and artificial intelligence. The aim of this study is to determine the role of digital traces generated during the educational process as a tool for digital transformation in educational management, as well as to demonstrate the practical implementation of digital trace processing using the VKontakte social network API and data analysis methods (Educational Data Mining). Materials and Methods . The theoretical and methodological basis for the study was formed by the work of Russian and international researchers in the field of digital data analysis arising during the educational process. The study utilized data analysis methods, natural language processing techniques, and Python libraries such as pandas, numpy, mathplotlib, and others. The empirical portion of the study is based on the analysis of the digital footprint of the educational organization's communities on the social network VKontakte, represented as unstructured text. Results . Research shows that large volumes of heterogeneous digital trace data, including those presented in the form of semi-structured data, inevitably arise in the context of the digitalization of education and ensuring the information openness of educational organizations. This data is of interest for educational analytics used to address issues related to the digital transformation of the educational process, the digital transformation of educational management, and the continuity and integration of educational levels. The digital trace generated through interactions with the electronic information and educational environment and other digital resources of educational organizations on the internet (websites, social media pages, and instant messaging apps) opens up opportunities for analyzing data on the educational process and participants in educational relationships. However, systematic approaches to its analysis and use in the context of the digital transformation of education are required, including those that take into account legal requirements for personal data, ethical aspects, and security aspects. This article examines the prospects for analyzing digital data in educational organization communities on social media using data analysis and machine learning methods and presents a practical example of data analysis in such communities on the social network VKontakte using an API. Conclusion . The obtained results can be used both for initial studies of digital footprint analysis and as a basis for developing a system for generating educational analytics. Practical application of the results will facilitate the digital transformation of educational management.

Background and context: Open datasets play a crucial role in three prominent research domains that intersect data science and education: learning analytics, educational data mining, and artificial intelligence in education. Researchers in these domains apply computational methods to analyze data from educational contexts, aiming to better understand and improve teaching and learning. Research scope and gap: Providing open datasets alongside research papers supports research reproducibility, fosters collaboration, and increases trust in research findings. It also provides individual benefits for authors, such as greater visibility, credibility, and citation potential. However, despite these advantages, the availability of open datasets and the associated practices within the learning analytics research communities, especially at their flagship conference venues, remain unclear. Goal and method: To address this gap, we conducted a systematic survey of publicly available datasets published alongside research papers in learning analytics domains. We manually examined 1,125 papers from three respected flagship conferences (LAK, EDM, and AIED) over the past five years (2020–2024). We discovered, categorized, and analyzed 172 unique datasets used in 204 publications. Results and contributions: Our study presents the most comprehensive collection and analysis of open educational datasets to date, along with the most detailed categorization. Of the 172 datasets identified, 143 were not captured in any prior survey of open data in learning analytics. We provide insights into the datasets’ context, analytical methods, use, and other properties. Based on this survey, we summarize the current gaps in the field. Furthermore, we list practical recommendations, advice, and 8-item guidelines under the acronym PRACTICE with a checklist to help researchers publish their data. Lastly, we share our original dataset: an annotated inventory detailing the discovered datasets and the corresponding publications. We hope these findings will support further adoption of open data practices in learning analytics communities and beyond.

Adaptive AI-driven learning systems personalize instruction by estimating learner state and dynamically selecting content, feedback, and pacing to improve mastery and engagement. This paper synthesizes peer-reviewed evidence on adaptive learning, intelligent tutoring, knowledge tracing, educational data mining, and recommender systems, and proposes an applied engineering framework suitable for deployment in higher-education STEM contexts. We ground personalization in classic student modeling (knowledge tracing) and modern sequence modeling (deep knowledge tracing), and integrate a multidimensional view of engagement to avoid reducing “engagement” to simple clickstream metrics. We then present a modular, service-oriented system architecture encompassing data ingestion, learner modeling, pedagogical decisioning, explainability, monitoring, and governance controls. A prototype evaluation is conducted using a simulation-based testbed (explicitly illustrative, not empirical) with synthetic learners and skills. Across 600 simulated learners and 25 skills over 120 learning steps, an adaptive policy improves average mastery (fraction of skills mastered at threshold) compared to non-adaptive paging and random sequencing, with markedly higher rates of reaching “80% mastery.” The results also show that naive optimization may widen outcome gaps across learner subgroups, motivating fairness-aware objectives and human-in-the-loop controls. Ethical, privacy, and accessibility requirements are addressed through risk management practices, differential privacy–compatible training options, transparent explanations, and WCAG-aligned interface design.

Because of the substantial class disparity and the intricate interactions between academic, behavioral, and socioeconomic characteristics, anticipating student academic performance and dropout rates continues to be a major issue for institutions of higher learning. To improve the dependability and credibility of multiclass student outcome prediction, this study suggests a strong, multi-objective, and uncertainty-aware predictive framework that combines the Random Forest (RF) classifier with Holistic Swarm Optimization (HSO). The suggested method creates a multi-objective optimization problem that simultaneously maximizes macro F1-score, controls model complexity, and lessens inter-class performance disparity. Thereby, the model promotes fairness across student outcome categories, in contrast to traditional optimization strategies that only concentrate on predictive accuracy. Furthermore, by utilizing ensemble-based probability dispersion, the framework integrates uncertainty-aware prediction, making it possible to identify high-risk students with different degrees of confidence to assist practical academic interventions. According to the results of experiments, the suggested HSO-RF framework greatly reduces the performance gap between outcome classes while achieving the best overall predictive performance, reaching an accuracy of 77.74%, a macro F1-score of 0.69, and a weighted F1-score of 0.76. The analysis shows that academic, socioeconomic, and administrative characteristics serve as significant markers of student motivation, stability, and vulnerability in addition to computational benefits. The suggested architecture advances appropriate and trustworthy educational data mining and offers a dependable decision-support tool for early warning systems.

Student performance evaluation at SDN 48 Kota Jambi has been traditionally conducted manually, which is inefficient and often subjective. This study aims to provide an objective classification of students’ academic achievement using data-driven methods. The research applies the Knowledge Discovery in Databases (KDD) framework, which involves data selection, preprocessing, clustering, and evaluation. The dataset consists of knowledge scores from 152 elementary students across seven subjects, obtained from the Merdeka Curriculum report cards. Data preprocessing included cleaning and normalization to ensure consistency. K-Means clustering was implemented using RapidMiner, with the optimal number of clusters determined through the Elbow Method. Cluster validity was assessed using the Davies–Bouldin Index (1.226) and the Silhouette Coefficient (0.245). The results produced three clusters: high achievers (30.9%), medium achievers (27.0%), and low achievers (42.1%). Centroid analysis indicated that Mathematics and Physical Education were the most discriminative subjects across groups. These findings highlight a substantial proportion of students requiring remedial intervention and support differentiated learning strategies. The contribution of this research lies in applying educational data mining techniques to an elementary school context in Jambi, integrating both quantitative indices and qualitative validation with teachers. The study demonstrates that clustering methods can enhance educational decision-making, providing a basis for adaptive teaching, targeted interventions, and resource allocation in elementary education.

Predicting students’ academic outcomes has become a central focus within Educational Data Mining (EDM) to support early academic interventions, minimize dropout risks, and promote personalized learning strategies. The availability of diverse educational data—ranging from demographic details and previous grades to behavioral engagement and digital learning activity—has encouraged the adoption of machine learning (ML) approaches for this purpose. This analysis synthesizes major ML techniques applied to student performance prediction, including regression models, decision trees, random forests, support vector machines, k-nearest neighbors, naïve Bayes classifiers, artificial neural networks, gradient boosting methods such as XGBoost and LightGBM, and deep learning models like CNNs and RNNs. Common evaluation measures—such as accuracy, precision, recall, F1-score, ROC-AUC, MAE, MSE, RMSE, and R²—are also examined. Despite their potential, challenges persist, including inconsistent data quality, complex feature selection, privacy and ethical concerns, limited interpretability, poor generalization across institutions, and the need for frequent model updates. Future research should emphasize explainable AI (XAI), temporal modeling techniques, integration of behavioral and psychological indicators, and transfer learning to enhance scalability and adaptability. Overall, this study highlights the promise of ML-driven systems in improving educational outcomes when combined with ethical, interpretable, and context-aware design principles.

Effective prediction of academic risk is vital in higher education to enable timely intervention and support student retention. While the introduction of Educational Data Mining (EDM) has enhanced prediction effectiveness, existing research often focuses only on single factors or large scale samples, and is notably deficient in providing transparent explanations for prediction results. To address these gaps, this study proposes an Explainable Artificial Intelligence (XAI) framework for predicting and interpreting academic risk within a high-dimensional, small sample context. Based on a dataset from a specific student cohort, we employed an ML model combined with SHapley Additive exPlanations (SHAP) method as the XAI framework. The findings provide two major contributions to the “Data-Related Challenges in ML” discussion. Firstly, by leveraging the XAI framework, it successfully enhances data interpretability, revealing the out-of-class peer support as the feature with the strongest association with academic risk, which is a complex and often underestimated data dimension, surpassing traditional academic metrics. Specifically, learning support from peers is identified as the most critical feature in mitigating risk at both the group and individual levels. Secondly, methodologically, this framework validates a reliable approach for extracting meaningful, trustworthy, and interpretable knowledge from limited and specific cohort data, offering a solution for applications with highly contextualized and precise interventions, where large, generalizable datasets are impractical. In conclusion, this study enhances the transparency and trustworthiness of ML in EDM, ensuring responsible intervention strategies in academic risk prediction.

Digital transformation in higher education is reshaping how institutions design and deliver their curricula, with a growing emphasis on student agency and personalized learning paths. This study employs educational data mining techniques to analyze student preferences and satisfaction with elective courses at Kryvyi Rih State Pedagogical University in Ukraine. We investigate patterns in course selection, satisfaction determinants, and the effectiveness of the university's individual educational trajectory framework among 1,089 students. Our analysis reveals four distinct student segments with varying preferences and satisfaction profiles. Information availability before selection, alignment with career goals, teaching quality, and course relevance emerge as significant predictors of student satisfaction. We propose a data-driven framework for optimizing elective course systems that incorporates learning analytics, personalized recommendation engines, and enhanced information platforms. This research contributes to understanding how educational technology can better support student agency in curriculum customization while addressing critical issues of accessibility, equity, and educational quality. The findings align with Sustainable Development Goal 4 (Quality Education) by promoting inclusive and personalized educational opportunities that prepare students for future employment challenges.

The advent of the Metaverse represents a significant paradigm shift in educational technology, transitioning from two-dimensional Learning Management Systems to immersive, three-dimensional virtual environments. While the pedagogical potential of these platforms is theoretically high, empirical evaluation of learning outcomes remains a complex challenge. This paper proposes a comprehensive framework for evaluating student performance and cognitive development within Metaverse-based educational platforms by leveraging Learning Analytics and Educational Data Mining. We investigate the correlation between immersive behavioral metrics—such as spatial trajectory, object interaction frequency, and gaze latency—and traditional learning outcomes. By applying clustering algorithms and predictive modeling to multimodal data streams generated within virtual environments, we identify distinct learner archetypes and engagement patterns. Our methodology addresses the limitations of self-reported surveys by introducing objective, data-driven indicators of engagement and knowledge acquisition. The findings suggest that while high immersion correlates with increased motivation, it introduces unique cognitive load challenges that must be managed through adaptive instructional design. This study provides a foundational rubric for educators and developers to assess the efficacy of immersive learning systems systematically.

Adaptive assessment engines represent a significant advancement in educational technology, integrating artificial intelligence with personalized learning to transform how academic evaluation is conducted in digital ecosystems. Reinforcement Learning (RL), with its capacity for sequential decision optimization, dynamic feedback processing, and autonomous policy refinement, provides a robust foundation for designing evaluation systems that adaptively tailor question difficulty, content progression, and diagnostic insights to each learner’s cognitive profile. Unlike static, uniform examinations, RL-driven assessment engines continuously observe learner behaviour, infer skill mastery, predict performance trajectories, and modify assessment pathways in real time to improve both accuracy and learning outcomes. This paper examines the theoretical underpinnings, algorithmic mechanisms, and behavioural implications of reinforcement learning in personalized academic evaluation, integrating insights from educational data mining, psychometrics, and intelligent tutoring system research. Through analysis of adaptive reward modelling, state–action representations, skill-mapping architectures, and policy optimization strategies, the study highlights how RL-based evaluation enables precision diagnostics, reduces test anxiety, enhances engagement, and supports mastery-based progression. The analysis further explores ethical, fairness, and transparency challenges, emphasizing the need for interpretable and bias-aware adaptive systems. The paper establishes a comprehensive foundation for understanding how reinforcement learning can advance personalized academic evaluation and shape the future of AI-enabled education.

This study optimizes the parameters of an Artificial Neural Network (ANN) to predict the academic performance of Automotive Vocational Education (AVE) students using a hybrid approach of Response Surface Methodology (RSM) and Genetic Algorithm (GA). Although ANNs have potential for educational data mining, parameter determination by trial and error reduces prediction accuracy. This study addresses this problem by applying the Central Composite Design (CCD) to identify influential parameters and their interactions, then refining them using RSM-GA. The subject of this research is Vocational High School (VHS) Antartika 1 Sidoarjo, East Java, Indonesia. The data include 884 students' academic records across 11 subjects from 2021 to 2023. The performance of the ANN is evaluated using the Mean Squared Error (MSE) as the primary criterion. Additionally, this study incorporates the Signal-to-Noise (S/N) Ratio from the Taguchi Method to assess model robustness. The S/N Ratio serves as an indicator of the ANN's stability in error minimization, applying the Smaller-the-Better (STB) criterion to optimize predictive accuracy. The results show the optimal ANN configuration: 3 hidden layers, 20 neurons, tansig-purelin activation function, trainbr algorithm, learning rate 0.001, and 300 epochs, with a Signal-to-Noise Ratio of 22.84 and a correlation coefficient of 0.89. The RSM-GA approach showed a 2.5% increase in the S/N Ratio compared to the CCD experimental method, making it more effective in identifying at-risk students early. These findings provide a systematic framework for ANN optimization in vocational education, with implications for the design of personalized learning interventions.

Decision tree models are widely applied in educational data analysis due to their simplicity and interpretability. However, these models exhibit high sensitivity to variations in training data, where minor sampling changes can result in substantially different tree structures, potentially reducing model reliability and consistency. This study aims to empirically investigate the structural robustness of decision trees under variations in educational data sampling and to evaluate strategies for improving model stability. An experimental framework is implemented using several educational datasets processed through random subsampling, bootstrap resampling, and k-fold cross-validation. Structural robustness is quantitatively assessed using tree edit distance, node similarity ratio, and tree depth variability. The results indicate that small sampling perturbations can cause significant structural divergence, particularly in datasets characterized by high noise levels and feature correlations. Nevertheless, pruning techniques and ensemble-based stabilization methods effectively enhance structural consistency and reduce model variance. These findings highlight the importance of robustness evaluation in educational data mining and provide empirical insights for developing reliable, stable, and interpretable decision-support systems in educational environments.

This study takes 300 literatures (from 2014 to 2024) included in the Web of Science Core Collection (mainly SSCI and SCI-E) as the data source and uses CiteSpace 6.3.R1 software to analyze the evolution of Profile Technology Applied in Education (PTAE). The research identified three core clusters: learner analysis and recommendation, intelligent technology-driven Educational Data Mining (EDM), and governance of the blended learning ecosystem. It reveals the deep transformation path of the field from basic feature characterization through data-driven prediction to a panoramic intelligent ecosystem. The research finds that the core research model is transitioning from static portrayal to a dynamic and precise intervention mechanism based on a panoramic learning analytics framework, establishing the underlying logic of modern adaptive systems Based on the findings, this paper explores algorithmic fairness and ethics challenges, and proposes practical strategies for building an ecologically adaptive intelligent educational environment.

Educational Data Mining (EDM) increasingly depends on large, high-quality datasets to drive predictive and adaptive learning systems. However, data scarcity, privacy restrictions, and limited accessibility severely hinder research reproducibility and cross-institutional collaboration. Synthetic data generation provides an emerging solution, enabling the creation of artificial yet statistically realistic datasets that preserve analytical utility while preserving student privacy. This study benchmarks four generative approaches, namely Gaussian Copula, CopulaGAN, Conditional Tabular Generative Adversarial Networks (CTGAN), and Tabular Variational Auto Encoders (TVAE), on student data from six undergraduate courses at a European university. Using the open-source Synthetic Data Vault (SDV) framework, we evaluate the fidelity and Machine Learning utility of synthetic student records through Random Forest classifiers across five metrics, namely accuracy, F1-score, precision, recall, and Area Under Curve (AUC). The results show that synthetic data can achieve 96–98% of the predictive performance obtained when training on real data, with TVAE consistently demonstrating the highest multivariate fidelity. Our contributions are threefold: (i) we introduce a reproducible benchmarking pipeline for synthetic data evaluation in educational settings; (ii) we empirically compare statistical and deep generative synthesizers on real-world tabular student data; and (iii) we identify critical research directions related to privacy and reproducibility. The findings position synthetic data generation as a foundational technology for ethical and privacy-preserving EDM.

This Systematic Literature Review (SLR) examines the application of Educational Data Mining (EDM) in online learning from 2015 to 2025 using the PRISMA approach. Thirty-two studies were analyzed to identify the data mining techniques used, the factors analyzed, and the extent to which the literature considers the equity and accessibility dimensions. The review results indicate that EDM is widely applied to predict academic performance, identify learning behavior patterns, detect at-risk students, and analyze the use of learning resources. The dominant techniques include classification, prediction, sequence analysis, process mining, and clustering. However, the equity and accessibility aspects are rarely discussed explicitly most studies only implicitly address accessibility through digital interaction behavior, while social factors related to equity, such as learning readiness, environmental support, and the digital divide, appear in only a small proportion. Furthermore, the variety of data formats and limited course coverage limit the generalizability of the findings. Overall, this study emphasizes the need for stronger integration between educational analytics and the social dimension for EDM to more effectively support equitable distribution of quality and access to online learning.

The article presents the results of an empirical study on the effectiveness of a digital platform for network projects in additional education for developing children’s inventive competencies. The pedagogical experiment (n = 250, 2022–2024) compared traditional teaching methods with the use of the online course “Nobel Heirs” and a learning analytics system. Torrance Test of Creative Thinking (figural battery), subjectivity questionnaires, and platform activity log analysis were used. Statistical processing included Student’s t-test, analysis of variance, and effect size calculation (Cohen’s d coefficient). Results showed statistically significant differences between experimental (n = 128) and control (n = 122) groups: fluency (t = 4.82; p < 0.001; d = 0.79), flexibility (t = 3.94; p < 0.001; d = 0.68), originality (t = 5.21; p < 0.001; d = 0.84), subjectivity in co-creation (t = 4.15; p < 0.001; d = 0.72). Correlation analysis revealed a significant relationship between platform activity and creativity growth (r = 0.64; p < 0.01). A metric for evaluating the effectiveness of the digital environment DICE (Digital Invention Creativity Effectiveness) was developed, considering product creation frequency, solution originality, project completion rate, and collaboration interactivity. The data demonstrate the feasibility of integrating educational data mining methods into children’s additional education practice.

The development of data analysis techniques and intelligent systems has had a considerable impact on education, and has seen the emergence of the field of educational data mining (EDM).The Early Warning System (EWS) has been of great use in predicting at-risk students or analyzing learners' performance.Our project concerns the development of an early warning system that takes into account a number of socio-cultural, structural and educational factors that have a direct impact on a student's decision to drop out of school.We have worked on an original database dedicated to this issue, which reflects our approach of seeking exhaustiveness and precision in the choice of dropout indicators.The model we built performed very well, particularly with the K-Nearest Neighbor (KNN) algorithm, with an accuracy rate of over 99.5% for the training set and over 99.3% for the test set.The results are visualized using a Django application we developed for this purpose, and we show how this can be useful for educational planning.

Multiple Educational Data Mining Approaches to Discover Patterns in University Admissions for Program Prediction

Feature selection plays a critical role in improving the efficiency, accuracy, and interpretability of machine learning models, particularly when dealing with high-dimensional datasets. Among various approaches, wrapper-based feature selection methods are known for their ability to capture feature interactions by directly optimizing model performance. This study presents a comprehensive comparative analysis of six wrapper feature selection techniques—Recursive Feature Elimination (RFE), Sequential Forward Selection (SFS), Sequential Backward Selection (SBS), Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Differential Evolution (DE)—in conjunction with five widely used classification algorithms: Decision Tree, K-Nearest Neighbour, Random Forest, Logistic Regression, and Support Vector Machine. Experiments are conducted on an educational dataset comprising 395 student records with 30 attributes obtained from the UCI repository, using different feature subset sizes (all features, top 20, top 15, and top 10). Model performance is evaluated using accuracy, precision, recall, F1-score, and AUC. The results demonstrate that wrapper methods significantly enhance classification performance while reducing dimensionality, with GA and RFE consistently emerging as the most effective techniques across multiple classifiers. DE also shows strong performance, particularly with Logistic Regression and Random Forest, whereas PSO generally underperforms in terms of AUC. Furthermore, reducing the feature set does not adversely affect predictive accuracy and, in several cases, leads to improved generalization. The findings confirm the effectiveness of wrapper methods for educational data mining and provide practical insights for selecting optimal feature–classifier combinations.