International Journal of Artificial Intelligence and Education Technology

The detection of students who may face academic difficulties or withdraw early gives universities a limited window in which to provide effective support. This study evaluates four machine-learning classifiers on the Open University Learning Analytics Dataset (OULAD), which contains data from 32,593 students across 22 distance-learning courses. The classifiers—logistic regression, decision tree, random forest, and gradient boosting—use demographic and contextual variables together with virtual learning environment (VLE) engagement features. VLE behaviour dominates Random Forest feature importance: total clicks, active VLE days, average daily clicks, and first-assessment score account for 92.8% of total importance. Random Forest achieves strong held-out test performance (AUC = 0.998, F1 = 0.978, accuracy = 98.2%), while the Decision Tree achieves an AUC of 0.959, illustrating the performance–interpretability trade-off. At-risk students generate an average of 49.0 VLE clicks, compared with 203.0 for not-at-risk students (t = 104.0, p < 0.001). The study describes an end-to-end prediction pipeline and discusses its implications for early-alert systems and the ethical deployment of predictive analytics in higher education.

Modern higher education campuses now use intelligent learning tools as standard educational resources, yet students’ learning results depend on their understanding of these tools and their implementation in academic work. The study analyzes how students prepare to use educational tools while investigating the connection between their preparedness and their judgment of educational benefits. The study uses an open student-perception dataset to conduct empirical research that includes developing constructs, profiling readiness, creating predictive models, and establishing pathways. It introduces two measurement methods: source breadth, which measures how students acquire knowledge about intelligent tools through different information channels, and an advantage score, which presents perceived benefits for educational activities. The three-profile segmentation method shows that different groups in the sample display distinct levels of preparedness and value assessment. The Random Forest model demonstrates superior performance because it achieves the highest accuracy among all tested models in the predictive stage. The selected model exhibits an accuracy rate of 0.789, a precision rate of 0.714, a recall rate of 1.000, an F1 score of 0.833, and an area under the receiver operating characteristic curve of 0.806 in hold-out evaluation. The analysis of variable importance indicates that AI knowledge, grade-point average, information breadth, and profile membership serve as the main factors that explain the results. The final stage transforms analytical results into distinct educational pathways that focus on developing essential literacy skills, implementing structured curriculum materials, providing support for governance matters, and enabling advanced collaborative learning. The results demonstrate that the educational benefits of intelligent tools depend more on students’ preparedness to use them than on their initial exposure to the tools.

Automated Essay Scoring (AES) technologies have been extensively researched for holistic, topic-specific scoring, but their use to predict multiple analytic writing quality traits of English Language Learner (ELL) student essays has received less attention. This research contributes to this knowledge gap by systematically investigating multi-trait AES on the ELLIPSE corpus (Learning Agency Lab, 2022), a publicly accessible dataset of 6,482 argumentative essays written by grades 8–12 ELLs and rated by human raters on six analytic traits: cohesion, syntax, vocabulary, phraseology, grammar, and conventions. We experiment with five models: Ridge regression, Support Vector Regression (SVR) with a radial basis function (RBF) kernel, Random Forest, fine-tuned BERT-base-uncased, and fine-tuned DeBERTa-v3-base. The mean Quadratic Weighted Kappa (QWK) across six traits is highest for DeBERTa-v3 (0.726), a 26.5-point improvement over the Ridge baseline (0.461) and a 6-point improvement over BERT (0.666). Phraseology is the most difficult trait to score automatically (DeBERTa QWK = 0.701) and cohesion the easiest (DeBERTa QWK = 0.742). Analysis of inter-trait correlations reveals high co-variation between vocabulary and phraseology (r = 0.79), which may reflect common linguistic skills that can be leveraged by multi-task learning. This research sets a replicable baseline for multi-trait AES on the ELLIPSE corpus and suggests that phraseology scoring is the most urgent area for future architectural innovation.

Student retention in higher education institutions is a critical problem that causes academic and financial challenges to individual students, schools, and entire countries. The present study compares five machine-learning classifiers—Linear Discriminant Analysis (LDA), K-Nearest Neighbours (KNN), Support Vector Machine (SVM), Random Forest (RF), and Gradient Boosting (GBM)—using data for 4,424 students from the Realinho et al. (2022) dataset. The dataset contains demographic, socioeconomic, macroeconomic, and academic-performance information from a Portuguese higher-education institution. Mutual-information feature selection reduces the 22-dimensional feature space to the 12 features with the highest discriminative power. Five-fold stratified cross-validation shows that the SVM with a radial-basis-function kernel achieves the best overall performance, with 97.1% accuracy and an F1-score of 0.954; all five models achieve an AUC greater than 0.981. The importance analysis reveals that four measures of academic success from the first two semesters account for 87.6% of the Random Forest feature importance, driven by the number of curricular units passed during the second semester (importance = 0.335). Socioeconomic, demographic, and macroeconomic features together contribute less than 13%. These findings provide empirical evidence about student-retention risk factors and inform the design of higher-education early-warning systems.

This study examines the theoretical, methodological, and applied aspects of integrating artificial intelligence (AI) into accounting systems and tax administration using expert evaluation methods. A systematic review of contemporary academic literature clarifies the conceptual foundations of AI and its role in the digital transformation of accounting and fiscal processes. The analysis focuses on the automation of accounting and managerial operations, intelligent accounting information systems, and virtual assistants for tax compliance and automated reporting. Selected applications, including Robotic Process Automation (RPA), accounting software robots (RobBee), and the DavrOn virtual assistant, illustrate the practical potential of intelligent technologies in financial and tax management. The results indicate that AI-based solutions can improve operational efficiency, data reliability, procedural transparency, and the adaptability of managerial decision-making as data complexity increases.