In this paper, we use the machine learning model to make predictions about the winning bid rate of apartments nationwide. The winning bid rate for apartments should consider various variables. There is a possibility that the existing hedonic priming models might predict uncertain results because of methodological constraints. In this paper, we aim to improve the predictions of apartment auction winning rates by utilizing algorithms such as Random Forest, XGBoost, LightGBM, and DNN, which are robust to problems such as nonlinearity and multicollinearity. A total of 111,232 nationwide apartment auction data were learned and tested from January 2010 to June 2020 by using the data provided by the GG auction and macroeconomic variables collected from KOSIS. In addition, a moving window methodology and an extending window methodology are applied considering by the characteristics of the social science data whose probability structure changes over time. Empirical study shows that the Gradient Boosting models outperforms other models in terms of MAPE, RMSE, MedAE, and AbsMean. There is no significant difference between a moving window methodology and an extended window methodology.