In this study, the packing density prediction model was established by the machine learning (ML) algorithm based on the drawing simulation to investigate the effect of design parameters of the multi-pass wire drawing processes on the powder density of superconducting MgB2 wire. In the drawing simulation, the modified Drucker-Prager Cap (DPC) model was used to capture the deformation behavior of the powder during the drawing processes, and the commercial finite element software ABAQUS was used to implement the modified DPC model. For the numerical generation of training and test data, the drawing simulations were conducted by changing the design parameters of the billet and drawing dies for the MgB2 drawing process. The drawing simulation results in terms of powder density influencing the superconducting properties of MgB2 wire were used as an output feature in ML. Of the output features, the rarely observed data were resampled by the Synthetic Minority Over-Sampling Technique for Regression with the Gaussian Noise algorithm [1]. R-squared cross-validation was adopted to find the best ML algorithm among artificial neural network, Gaussian processing regression, random forest, and support vector regression. The prediction model was used to investigate the effect of design parameters on the model output, in which the Shapley additive explanations method [2] was used to account for the feature effects on model outputs.