Abstract

Modern distributed systems, cyber‑physical infrastructures, and mission‑critical applications face frequent failures, performance degradation, and unpredictable dynamic conditions. Traditional fault‑tolerance approaches such as redundancy and reactive recovery are insufficient for next‑generation systems where service continuity and minimal downtime are required. Self‑healing system architectures integrate predictive failure modeling with proactive recovery strategies to detect impending anomalies and automatically trigger corrective actions before failure manifestation. These architectures leverage health monitoring, data analytics, machine learning, model‑based reasoning, and control mechanisms to enable system components to anticipate faults, isolate degraded elements, and reconfigure resources in real time. Predictive models—built using time‑series analysis, statistical learning, and failure pattern recognition—estimate the likelihood of component degradation, enabling early warnings and automated interventions. Proactive recovery strategies include automated checkpointing, dynamic resource redistribution, micro‑restarts, automated migration, and adaptive reconfiguration to prevent systemic collapse. This review synthesizes current approaches to predictive failure modeling and self‑healing architectures, highlighting strengths, limitations, and integration challenges. Experimental evaluations show that predictive models significantly improve availability, reduce downtime, and enhance system resilience compared to purely reactive systems. The study underscores the need for adaptive, hybrid techniques combining machine learning with formal models and outlines future research directions for scalable, intelligent self‑healing infrastructures.