- Radkevych, Olexandr (orcid.org/0000-0002-2648-5726) (2026) The potential of artificial intelligence in educational analytics and managerial decision-making Professional pedagogics, 2 (33). pp. 168-184. ISSN 2707-3092
![[thumbnail of Oleksandr Radkevych.pdf]](https://lib.iitta.gov.ua/style/images/fileicons/text.png "Oleksandr Radkevych.pdf") |
Text
Oleksandr Radkevych.pdf Download (1MB) |
Abstract
Relevance. Large-scale national and international assessments generate substantial volumes of performance, contextual, and administrative data; however, their use in education governance is often confined to a retrospective description of mean scores and between-group gaps. The need to move from descriptive reporting to anticipatory, differentiated, and evidence-based governance brings to the fore the application of artificial intelligence (hereinafter – AI) technologies in educational analytics. At the same time, the high legal and social stakes of educational decisions call for overcoming technological determinism, preventing discriminatory profiling, strictly complying with personal data protection legislation, and guaranteeing human primacy and accountability. Purpose. To substantiate a conceptual model for transforming the results of national and international assessments into managerial decisions using AI tools, and to determine the organisational, managerial, methodological, ethical, and legal conditions for its responsible implementation. Methods. The study employs: conceptual analysis to distinguish educational analytics from static data visualisation, to differentiate four levels of analytics (descriptive, diagnostic, predictive, and prescriptive (recommendation-oriented)), and to clarify the essence of the concepts of “algorithmic fairness”, “differentiated explainability”, and “socio-technical system”; comparative analysis to juxtapose the purposes, scales, target populations, and limits of normative interpretation of assessment data (PISA, TIMSS, the National Multi-Subject Test (NMT), nationwide monitoring studies, and digital traces from educational platforms), and to weigh the performance of complex algorithmic models against interpretable baseline statistical models; thematic coding to systematise the provisions of international framework documents (OECD, UNESCO, EU legal acts) and scholarly works in order to identify key risks (sampling bias, temporal degradation of predictive models, bias against particular groups of learners, and over-reliance on automated decisions) and the potential of algorithmic methods in education quality assessment; critical synthesis to reconcile the standards of educational measurement (test measurement error, equivalence of test booklet forms, and the probabilistic nature of scores) with the analytical potential of machine learning methods and language models and with the legal norms governing the protection of confidential information; and structural-functional modelling to design a closed cycle of managerial decision support (data – algorithmic model – expert verification – managerial decision – impact monitoring), to substantiate the component structure of evidence-based education quality governance, and to formalise a composite criterion of the fitness of analytical models (Q). Results. A fourteen-component model for transforming the results of national and international assessments (PISA, NMT, monitoring studies) into managerial actions is substantiated. The model is built as a closed cycle: from specifying the educational construct (a construct passport) and ensuring the interoperability of heterogeneous registers and reference directories to expert verification of signals and evaluation of the actual consequences of managerial actions. A multi-criteria composite index of model fitness (Q) is developed, which weighs validity with respect to the educational construct, robustness of algorithms on new samples, algorithmic fairness, intelligibility for users, and the aggregate cost of possible errors. Scenarios of AI application are systematised according to the levels of education governance (national, regional, institutional, and the level of the individual learning trajectory), and information security requirements are defined to preclude direct access of generative models to learners’ personal data. Conclusions. It is demonstrated that AI in educational analytics serves to augment expert judgement rather than to act as an autonomous decision-maker; an automated output is a probabilistic signal for a professional, not an imperative directive. The transition from prediction to action requires an independent evidence base on the effectiveness of pedagogical interventions and pre-agreed criteria for evaluating possible errors, where the unconditional priority is to prevent discriminatory profiling of students and to protect comprehensively the rights of all participants in the educational process. For the education system of Ukraine, a phased transition to AI-supported decision-making is essential: from ensuring the interoperability of the core classifiers and reference directories of state information systems (in particular, AIKOM and EDEBO) and implementing low-risk pilot projects (automated processing of response datasets, data validity checks, and error detection) to legally regulated predictive modelling subject to mandatory interdisciplinary audit, guaranteed transparency of conclusions, and the right to appeal and to mandatory human review.
Downloads
Downloads per month over past year
Actions (login required)
 |
View Item |