Интерпретируемая нечёткая модель количественной оценки учебной вовлечённости по поведенческим данным цифрового следа
Year: 2026
Type: article
Abstract: Introduction. The study develops and empirically evaluates an interpretable fuzzy model for assessing learning engagement from behavioural digital footprint data. The construct and its measurement are explicitly separated: learning engagement is treated as a latent construct not directly observable, while the model assesses its observable behavioural projection in the digital footprint – the applied-effort coefficient. The problem is motivated by the ceiling effect: half of the active students earn the maximum score on every mandatory assignment, so formal grades no longer differentiate engagement levels. Materials and Methods. The model maps observable projections of constructs from Self-Determination Theory, Expectancy-Value Theory and the ARCS model onto three digital footprint channels – competence weighted by cognitive complexity, optional advanced tasks, and subject-relevant extracurricular activity rated independently by two experts – aggregated by a Mamdani fuzzy inference system with an explicit competence gate that disallows compensating for unmet coursework with activity. Results. In an experiment on real de-identified records of 62 students of a Database Theory course, the applied-effort coefficient spreads the cohort compressed against the grade ceiling and shows criterion validity against two independent outcomes: the final examination and the grade-point average across other courses; the proactivity channels predict the examination beyond coursework performance. Discussion and Conclusion. In terms of predictive accuracy, the model is not statistically distinguishable from logistic regression on the same channels – the confidence intervals overlap, and no accuracy advantage is claimed. An ablation analysis shows the competence gate suppresses the “activist without foundations” profile that a linear index misjudges as satisfactory; the fuzzy boundary removes the sharp discontinuity at borderline competence. The model’s contribution lies in interpretable non-linear behaviour rather than predictive gains. The model constitutes a scalable methodological framework: course-specific parameters are localized in explicitly described components and can be re-tuned for another course without changing the inference architecture.
Institution: MIREA - Russian Technological University
Language: English
Subfield: Computer Science Applications
Field: Computer Science
Domain: Physical Sciences
SDG: Quality education
Open Access status: diamond
OpenAlex ID: W7217455245