Implementation of computational fluid dynamics (CFD) for design of tactical missile systems is a well-established method in the literature. However, this methodology, being computationally expensive, leads to a significantly high amount of time when run-in combinations with iterative algorithms such as the evolutionary optimization. The idea of replacing these CFD models with machine learning–based surrogate models, if possible, can be highly useful in this regard. Among many such machine learning models, artificial neural networks (ANNs) are shown to be promising candidates for capturing nonlinearities involved in the first principles–based models (FPMs). However, heuristic-based determination of hyperparameters in ANNs became a barrier toward their successful implementation as surrogates. In this work, authors propose a novel multiobjective evolutionary optimization approach that aims to achieve optimal estimation of such hyperparameters: architecture, sample size for training, and choice of activation, simultaneously at the time of building ANN surrogates. The data for training the ANN models is obtained from the high-fidelity time-expensive CFD simulations for modeling the supersonic-flow over a cruciform missile system. The optimal surrogate models, built with an accuracy of 99%, are then used to generate the predictions for performing uncertainty analysis of individual effects and interaction effects of three design parameters: flux type, discretization order, and models for turbulence in ANSYS FLUENT on three variables of interest: the drag coefficient, the lift coefficient, and the coefficient of rolling moment of the missile using analysis of variance (ANOVA). © 2023 Elsevier Inc. All rights reserved.