This study presents a model to predict the employment status and the employment quality of college graduates using the random forest method of machine learning. In traditional regression analysis, there are some constraints on the problem of endogeneity and distribution of errors. However, the machine learning approach is relatively free from these constraints. The predictors used in this study include not only the objective characteristics of college graduates but also characteristics of respondents' subjective evaluation, including participation in employment programs, consideration of job selection, and emotional frequency. The estimation results show that the models using both objective and subjective predictors have better predictive performances than the model using only objective predictors in both the employment status and the employment quality models. As a result of analyzing the relative importance of predictors using the random forest method, not only the subjective variables such as householder status, parental cohabitation, and major, but also subjective variables, such as the emotional frequency, have an important effect on the employment status. This is also true of the model for predicting the quality of employment.