Neural Network Method for Predictive-Adaptive Stability Control of the Main Coordinator in Fog/Edge Infrastructures

Abstract

This article presents the SENTRY-C neural network method for predictive-adaptive stability control of the main coordinator node in distributed Fog/Edge infrastructures. Unlike classical intrusion-detection and risk-assessment approaches that operate in a reactive mode, SENTRY-C enables continuous forecasting of coordinator degradation and provides proactive control actions under dynamically changing load and threat conditions. The method integrates a Long Short-Term Memory (LSTM) recurrent neural network into the coordinator control loop to model temporal dependencies and evaluate the probability of failure. The experimental evaluation demonstrates that the use of LSTM improves short-term prediction accuracy of stability indicators by 25–35% and reduces the error of predicting degradation probability by up to 30% compared with baseline and basic RNN models. High Pearson correlation coefficients (0.90–0.94) confirm the reliability of the predicted parameters under rapid load variation, gradual degradation and short-term risk spikes. The results show that SENTRY-C maintains the continuity and robustness of coordinator functioning through adaptive parameter updates and real-time feedback. Future research is focused on integrating federated learning and reducing computational complexity to enable deployment of the method in resource-constrained Fog/Edge nodes