표준주택공시가격 적정성 제고를 위한 기계학습적 접근

Abstract

The paper deals with how to enhance the appropriateness of the posted prices of the standard houses with the perspective of machine learning. We formulate a classification problem with the binary response that is defined by the magnitude of the rate of posted price to actual price of the standard houses in each metropolitan city and province. Several variables regarding characteristics of the houses are used as predictor variables in the statistical modeling. The considered models are logistic regression, decision tree, bagging and gradient boosting. The performance comparison using ROC curve or lift charts suggests the gradient boosting as the best model in this situation. The modeling result can be utilized for adjusting the posted prices of standard houses in advance which leads to a better balanced distribution of the posted prices in terms of COD (coefficient of dispersion) and high rate of reflecting the actual prices to the posted prices. We analyses a real data set regarding posted prices and actual transaction prices and the result show that the machine learning approach can be effectively applied to enhance the appropriateness of the posted prices of standard houses.