• Linginedi Ushasree
• S. G. Shanthi
• Haritha Donavalli

Operators now run networks that are software-defined, virtualised, spread across cloud and edge sites, and expected to keep services alive with little manual intervention. That flexibility comes at a price. Faults travel along dependencies that are hard to see, alarms arrive in bursts, and the rule-based self-healing logic inherited from earlier generations usually reacts only after users have already noticed the problem. This paper reviews work published between 2020 and 2026 on applying artificial intelligence to the detection, prediction, diagnosis and repair of network faults. Five lines of research are examined in detail: graph neural networks (GNNs), which encode topology and dependency structure; Transformer models, which capture long-range temporal behaviour in telemetry; reinforcement learning (RL), which learns recovery and rerouting policies; explainable AI (XAI), which makes automated decisions auditable; and network digital twins (NDTs), which allow a candidate action to be tested before it touches live infrastructure. Related work on zero-touch network and service management (ZSM), alarm root-cause analysis and resilience-aware routing is also covered. Our reading of the literature is that each family of methods fixes a specific weakness of conventional operations, yet almost every published system addresses only one or two stages of the healing cycle. Very few connect perception, forecasting, causal diagnosis, action selection, validation and explanation in one loop. Building on this observation, we outline NeuroHeal-Net, a reference architecture that chains these functions together under explicit policy constraints, and we propose an evaluation protocol that measures service impact and safety rather than detection accuracy alone. We argue that the next generation of self-healing networks should be predictive, explainable, risk-aware and continuously validated.

self-healing networks; autonomous network management; graph neural networks; Transformers; deep reinforcement learning; explainable AI; network digital twin; root-cause analysis; zero-touch management; network resilience

"AI-Driven Self-Healing Computer Networks: A Review of Graph Neural Networks, Temporal Transformers, Reinforcement Learning, Explainable AI and Network Digital Twins ", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.13, Issue 10, page no.b682-b691, October-2026, Available :http://www.jetir.org/papers/JETIR2610167.pdf

"AI-Driven Self-Healing Computer Networks: A Review of Graph Neural Networks, Temporal Transformers, Reinforcement Learning, Explainable AI and Network Digital Twins ", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.13, Issue 10, page no. ppb682-b691, October-2026, Available at : http://www.jetir.org/papers/JETIR2610167.pdf