A Study on the Estimation of Apartment Rent Index Using Machine Learning

This study explored a method for constructing a rent index that systematically reflects structural characteristics of South Korea’s monthly rental market. Using actual transaction data, the study developed rent indices based on the repeat-sales model and the machine learning models (Gradient Boosting, Artificial Neural Network and Ensemble) and compared their respective outputs. Median points of the actual transaction data showed that monthly rent prices increased by approximately 1.52 times between January 2011 and December 2024. Both the repeat-sales-based and the machine learning-based indices exhibited similar cumulative growth. The former was prone to instability in periods without any transaction or in small-area units due to a reduction in sample size, whereas the latter demonstrated the ability to respond sensitively to abrupt market changes and reliably captured trends even during transaction gaps by incorporating nonlinear relationships and a wide range of input variables. The GBM-based index specifically showed strength in reflecting short-term fluctuations, while the ANN-based index, through continuous weight updates, had excellent smoothing effects and was advantageous for identifying long-term trends. The ensemble index, combining both models, offset each one