기계적 학습을 이용한 주택가격 예측

Abstract

This study compares the predictive power of a linear regression model, a big data analysis methodology, gradient boosting, and a support vector machine. The dependent variable was the housing price, and the independent variables were the housing Jeonse price, consumer price index, interest rate and the status of the building start. The time range was from January 2006 to July 2020. The spatial scope was the whole country, the metropolitan area, provinces and Seoul. As a result of empirical analysis, the gradient boosting model showed the lowest mean square error (MSE) value in all regions, indicating that it has the best predictive power. Looking at the influence of the variables of the gradient boosting model, it was found that the influence of the Jeonse price was the highest in Seoul and the metropolitan area, while the influence of the Consumer Price Index was higher than the Jeonse price in the nation and provinces. Looking at the conditional effect, although there are regional differences, in most regions, the consumer price index and the housing jeonse price showed a positive (+) effect on the housing sale price, and the effect of the construction start status was very insignificant. Also, interest rates were found to have a negative (-) effect on the housing sale price.