In this brief, we suggest machine learning (ML) approach for statistical random grain boundary (RGB) variation model of polysilicon channel in 3-D NAND flash memory. In order to analyze RGB variation, which greatly affects characteristics of devices, TCAD simulation is usually used, but the computational cost is too high to deal with a lot of RGBs generated in the polysilicon channel of numerous strings. Therefore, we devised an ML approach that can predict the effect of RGB variation on device characteristics and performances with high accuracy and efficiency. The presented model is constructed using artificial neural network (ANN) and has multi-input multi-output (MIMO) structure to apply complex data. The model was proven to be accurate by showing a small error of 3%–7% compared with TCAD simulation, and the efficiency was shown through 50 000 strings prediction in a short time using this model.