SYNTESIS OF COMPOSITE NEURAL NETWORK ARCHITECTURES FOR PREDICTIVE DATA PROCESSING IN DIGITAL INFOCOMMUNICATION EDUCATIONAL ENVIRONMENTS
Information and communication technologies, electronic engineering · 2026-06 · 6 (1) · 75-85
DOI: 10.23939/ictee2026.01.075 ↗Kurz gefasst
Fachartikel in der Zeitschrift Information and communication technologies, electronic engineering, 2026-04. Verfasst von Y. Makohon und 1 weiteren Autor:innen. Die Arbeit nennt 26 Literaturangaben.
Abstract (Originalsprache)
This study substantiates the methodological and technical principles for synthesizing composite neural network architectures designed for predictive data processing within digital infocommunication educational environments. In the context of global digital transformation and escalating infrastructural challenges, particularly in Ukraine, the research focuses on developing adaptive systems capable of operating efficiently on resource-constrained hardware. The core scientific novelty lies in the formalization of a three-stage structural integration mechanism facilitating a bidirectional information flow between recurrent (LSTM) and graph-based (GCN) components. This synthesis ensures that temporal behavioral patterns are processed in direct correlation with the non-linear topological structure of the Knowledge Graph. The methodological framework is validated using a longitudinal dataset comprising 1.28 million interaction events from 15,640 unique student profiles, integrated from the Open University Learning Analytics Dataset (OULAD) and localized Moodle LMS logs from Kharkiv National University of Radio Electronics. The study addresses a binary academic risk classification task with a 7-day forecasting horizon, utilizing 24 behavioral event types and a Knowledge Graph of 1420 competencies. Experimental results obtained on consumer-grade hardware with a 1 GB VRAM limit demonstrate a 14-fold reduction in GPU inference latency and a 95% decrease in video memory consumption compared to monolithic Transformer baselines. Furthermore, the convergence analysis reveals that the composite architecture achieves a stable performance plateau by epoch 15, significantly mitigating gradient volatility observed in memory-starved transformers. Beyond technical optimization, the paper proposes a theoretical framework for a multi-component loss function aimed at balancing accuracy, student privacy, and algorithmic fairness. The results confirm that the composite approach provides a viable technical foundation for maintaining cognitive resilience and bridging educational gaps in crisis-affected environments where reliance on centralized cloud infrastructure is constrained.
Zitieren
APA
Makohon, Y., & Rudenko, O. (2026). SYNTESIS OF COMPOSITE NEURAL NETWORK ARCHITECTURES FOR PREDICTIVE DATA PROCESSING IN DIGITAL INFOCOMMUNICATION EDUCATIONAL ENVIRONMENTS. Information and communication technologies, electronic engineering, 6(1), 75-85. https://doi.org/10.23939/ictee2026.01.075
Vancouver
Makohon Y, Rudenko O. SYNTESIS OF COMPOSITE NEURAL NETWORK ARCHITECTURES FOR PREDICTIVE DATA PROCESSING IN DIGITAL INFOCOMMUNICATION EDUCATIONAL ENVIRONMENTS. Information and communication technologies, electronic engineering. 2026;6(1):75-85. doi:10.23939/ictee2026.01.075
BibTeX
@article{Makohon2026,
author = {Makohon, Y. and Rudenko, O.},
title = {SYNTESIS OF COMPOSITE NEURAL NETWORK ARCHITECTURES FOR PREDICTIVE DATA PROCESSING IN DIGITAL INFOCOMMUNICATION EDUCATIONAL ENVIRONMENTS},
journal = {Information and communication technologies, electronic engineering},
year = {2026},
volume = {6},
number = {1},
pages = {75--85},
doi = {10.23939/ictee2026.01.075}
}
EUVIMED