Adaptive failure prediction in information systems based on ontological-entropy modeling

DOI: 10.31673/2412-9070.2025.045282

Abstract

This article presents a method for adaptive failure prediction in computer networks based on the integration of ontological analysis and entropy-driven modeling. An integrated model combining network structure, behavioral templates, and entropy characteristics is developed for real-time risk estimation and proactive response. Particular attention is given to the construction of an ontology capable of dynamic updates of concepts and properties reacting to network events. An adaptive entropy coefficient (AE) is introduced, combining classical Shannon entropy with contextual informativeness derived from the ontology. The model not only quantitatively evaluates system instability but also adapts risk assessment rules based on the operational environment.
Risk escalation scenarios were simulated using telemetry from IoT devices and events from SCADA system logs. It was found that AE values below 0.5 correlate with a failure probability exceeding 90% within the next 20 minutes. The model supports the construction of a dynamic event-dependency topology, enabling the detection of cascading threats and architectural vulnerabilities. The proposed approach provides multi-component stability assessment by integrating description logic, context-aware weighting, and adaptive processing of high-frequency data streams. The relevance of the study is confirmed by growing demands for proactive monitoring under highly dynamic infrastructure conditions.
The results demonstrate that the proposed model delivers high accuracy in detecting critical states, significantly outperforming classical statistical methods. Implementation scenarios were analyzed for energy, logistics, and telecommunications systems, considering computational resource constraints. A dynamic parameter recalibration algorithm was proposed, allowing the model to adapt to system reconfiguration without retraining. Thus, the ontological-entropy model offers strong scalability, autonomy, and practical applicability in critical environments.

Keywords: adaptive entropy; ontological modeling; failure prediction; information system; IoT; risk; anomaly; SCADA; cyber resilience.