Achieving high interception accuracy in Integrated Guidance and Control (IGC) for Autonomous Aerial Systems (AAS) is hindered by modeling uncertainties and system nonlinearities. While classical controllers like PID and LQR struggle with these dynamics, advanced robust controllers such as Sliding Mode Control (SMC) suffer from practical issues like chattering. This study investigates if a hybrid predictive control architecture can surpass both classical and robust nonlinear controllers in uncertain engagement scenarios. We hypothesize that augmenting a Model Predictive Controller (MPC) with an online neural-network (NN) based system identifier will enable adaptation to real-time dynamics, leading to superior performance. We propose an MPC-NN framework where an NN performs online parametric identification from sensor data, continuously updating the MPC’s internal prediction model at each sampling instant. The framework’s performance was rigorously benchmarked against PID, LQR, and a robust SMC baseline in 3D short-range air-defense scenarios using comprehensive Monte Carlo simulations. The proposed MPC-NN demonstrated decisive superiority. Under stochastic uncertainties, it achieved a mean terminal miss distance of 0.87 m, significantly outperforming SMC (1.98 m), LQR (7.91 m), and PID (18.45 m). The MPC-NN also proved more efficient, reducing interception time by 17% compared to SMC while consuming nearly half the peak control effort and completely avoiding actuator saturation—a problem prevalent in other controllers. These results validate our hypothesis, confirming that the adaptive MPC-NN architecture provides a transformative advantage in accuracy, speed, and efficiency. This framework represents a highly promising and practical solution for next-generation air-defense systems facing unpredictable, high-maneuverability threats.