New network infrastructures are strongly dynamic, programmable, and scalable due to extensive adoption of NFV and SDN. Higher operational complexity is traded off against these benefits, opening up the networks to anomalies like attacks, misconfiguration, and performance degradation. Traditional fault management and monitoring protocols are reactive in nature and involved human intervention, leading to added downtime and operational expense. For addressing such problems, this paper proposes an intelligent self-healing and anomaly detection framework in virtual networks through AI-based models and Java-based orchestration techniques. The anomaly detection engine utilizes a hybrid scheme that precisely identifies known and unknown anomalies by utilizing deep autoencoders for reconstruction error along with isolation forests for effective unsupervised scoring. Self-healing module works in conjunction with SDN/NFV controllers to perform corrective actions like scaling, migration, and rerouting, and root cause analysis (RCA) detects malfunctioning network services by analysing dependency graphs and applying probabilistic reasoning. 55% mean time to recovery (MTTR) decrease, fewer false alarms, and higher detection accuracy (>95%) are illustrated through experimental outcomes. This paper enhances resilient, flexible, and self-managed virtual network establishment.