For the forming of hollow components, cold radial forging (CRF) is thought to be one of the best manufacturing processes. However, minute modifications to the alloy composition during the stage of cold deformation processing can have an impact on the material's manufacturability. In order to tackle the issues of elevated optimization costs, challenging data acquisition, and forming quality prediction when combining multiple processes, like heat treatment and material composition. In this paper, we propose to build an intelligent prediction model to realize the prediction of the forming quality of CRF of 20CrMnTiH alloy using a back-propagation neural network (GA-BP) optimized by genetic algorithm. The evaluation of forming quality is based on damage, residual stress, and equivalent strain. The effects of alloy composition and spheroidal annealing (SA) process parameters are studied in relation to forming quality. The three prediction models of GA-BP perform better in terms of R2 (coefficient of determination) and MSE (mean square error) than the conventional BP neural network and four other machine learning techniques (gradient-enhanced decision tree, random forest, support vector regression, logistic regression, and logistic regression). Based on a thorough comparative analysis of the evaluation metrics of all three prediction models and multi-objective optimization, it can be concluded that the Pareto solution set produced by NSGA-II has the best distribution uniformity. Remaining stresses were reduced by 12 percent, equivalent strain by 15 percent, and component damage by 30 percent after the obtained process parameters (0.17%C-0.24%Si-0.81%Mn-0.03%P-0.03%S-1.15%Cr-0.05%Ni-0.06%Ti, Te: 60℃, ATi: 4.57h) were optimized using the suggested methodology. The findings demonstrate that the approach presented in this paper improves the CRF molding quality effect in addition to making the optimized model significantly more accurate at predicting molding quality precision.