Modelling of flow forming of tube-shaped solution annealed H30 Aluminium alloy is considered in the present study. Initially, a total of 136 experiments have been conducted to realize the process and subsequently influences of three inputs (feed–speed ratio, roller infeed and axial stagger) on the three outputs, viz. internal diameter, springback and ovality have been studied. Three neural network-based approaches (back-propagation neural network, limited-memory BFGS network and genetic neural system) have been developed for forward as well as reverse modelling of the process. During forward modelling, the performances of the three neural network-based approaches have been compared with the regression model. It is seen that GANN has performed much better compared to the other methods. Percentage accuracy in predicting ovality using regression analysis is the worst, and it necessitates consideration of more input process parameters for better prediction accuracy. However, NN-based approaches adapted such cases well. Comparison of all the three NN-based approaches among themselves has been made during reverse modelling. During this process, prediction accuracy, using LBFGSNN, is found to be better than the other two methods. Thus, it is perceived that NN-based models might suit better for prediction of shape accuracy of flow-formed shell.