Machine learning algorithms have been widely used in the prediction of groundwater level, but most of these models are incapable of screening and analyzing the types of input variables, and fail to further explore the nonlinear interactions between these input variables and the degree of their influence on groundwater level. To solve the above-mentioned problems, a GRA-FA-SVM hybrid model, which combines the support vector machine (SVM) with the grey relational analysis (GRA) and the factor analysis (FA), was established to predict the groundwater level. The relevant field observation data at two observation wells in Minqin County of northwestern China was used for validating the proposed hybrid model. The investigated results verify the effectiveness of the hybrid model and show that the GRA-FA-SVM model gives the highest accuracy.