Predicting the Apartment Transaction Price by Using Machine Learning Methods and the Fixed Effects Model

Abstract

Predicting the prices of housing is a critical issue for all real estate market participants, including financial institutions and governments, as well as homeowners and potential buyers. To evaluate the performance of a housing price prediction analysis model, this study compares machine learning methods and the fixed effects model based on the panel data of actual transactions between January 2006 and July 2020 in Seoul. For this purpose, we are also considering the characteristics of the apartment complex, based on the hedonic pricing approach, as well as macroeconomic variables and policy variables. We find that the Random Forest (RF) model shows higher predictive accuracy than the Multivariate Adaptive Regression Splines (MARS) model in machine learning. Our study shows the potential of the wide use of the machine learning method by utilizing micro-data in predicting and forecasting housing prices more accurately.