Deep Learning Application on Mobile Ad Hoc Networks under Malicious Attacks. This study explores the application of deep learning to Mobile Ad Hoc Networks (MANETs) under the influence of malicious attacks. More specifically, it simulates a total of 50 mobile nodes in an area of 100 × 100 and models the node mobility, contact probability, and communication links using a feedforward neural network. The system combines several performance metrics, including throughput, packet delivery ratio (PDR), and end-to-end delay, to allow you to see how the network behaves under both dynamic and adversarial conditions. We then develop a mathematical model to characterize dynamic topology and intensity of attack and network features respectively, paving the way to a deep learning-based surrogate model to predict on demand performance in real-time. Also, we have a utility optimization layer that balances between efficiency and resilience, which favours better utility for society in real world deployments. The results of the experiment exhibited decent performance with an average throughput of 10.9 pkts/s, average delay of 0.0504 s, and a 10.9% PDR. This deep learning showcased not only its efficacy in modelling and evaluating both MANET operations in evolving cyber threats and complex mobility patterns but also creating new solutions for optimizing such operations.