<p>Theself healing networks effectively monitor the real time manufacturing system in mobile networks to find the faults in the system. One of the main issues of the optical network are reliability, high data transfer rates, network performance, volume and cost efficiency. The traditional procedures contain many difficulties during fault prediction. It is necessary to find a solution to the fault detection. A novel fault detection framework in self healing network is implemented based on different deep learning models. Essential data for the analysis is taken from benchmark resources. Next, the fault tolerance data is subjected to the self healing network, the process recovered by automated triggered series of actions. In the self healing network, imbalance data is cleared by utilizing the Eigen-Entropy Synthetic Minority Oversampling Technique (EE-SMOTE), which is an oversampling model designed based on information entropy to offer better support for imbalance classification and also SMOTE is utilized to achieve the data balances for effective classification. Further, the cleared imbalance data is subjected to a Multi-scale Dilated Bidirectional Long Short-Term Memory with Attention Mechanism (MDBiLSTM-AM)-based fault classification phase. The developed fault detection framework in a self-healing network is given a better efficacy rate than the conventional methods in different investigational observations.Finally, the developed method is validated with several performance measures and compared with different traditional approaches. The developed MDBiLSTM-AM approach attains a maximum accuracy rate of 97.4, 97.44, and 97.46% for Dataset 1, Dataset 2, and Dataset 3. Thus, it demonstrates the developed method’s overall better performance than the existing methods.</p>