<p>In the field of cyber-security, the challenges related to cyber-attacks over MANET and growing sophistication have led researchers to accept predictive and reactive approaches. For categorizing packets as malicious or standard, the presented paper developed an intrusion detection system based on signature by employing deep hybrid network classifiers. Furthermore, an anomaly-based intrusion detection system is designed to recognize the anomalies by employing two pre-trained network models, i.e. VGG-19 and VGG-16 in the Hybrid Deep Convolutional Network over the MANET environment. The signature generator will process these anomalies. The training dataset is incorporated with relevant attributes extracted from a signature generator. Therefore, by including new attack models, this integration makes continuous learning by the hybrid model possible. The hybrid model’s self-healing property allows for early and quick attack discovery. It is done by generating attack features for unknown attacks that occurred previously. The UNSW-NB-15 dataset is utilized to assess the projected methodology, and the provided dataset is employed to evaluate the effectiveness of the projected hybrid classifier. When the experimental outcomes are compared to the conventional models, the proposed model also exhibits improved detection rates for unknown and known attacks.</p>