기계학습과 XAI를 활용한 아파트 가격과 지역특성과의 관계 분석

Abstract

This study aims to explain the relationship between variables that were not explained in the existing machine learning methodology by interpreting the relationship between apartment prices and regional characteristic variables using the XAI methodology. Furthermore, the study's goal is to collect various regional variables (population, occupation, income/consumption) in the basic area unit (postcode unit) that could not be addressed in previous domestic studies due to data collection difficulties. In terms of the independent variable, a dataset was created for each facility, population, occupation, income/consumption, and the dependent variable. The relationship between independent and dependent variables was inferred using a random forest among XGBoost and Bagging models and SHAP models among XAI methodologies on the constructed dataset. The spatial scope of the analysis was determined by dividing Seoul Metropolitan Government into national basic zone units, the content range was apartment among houses, and the period was empirically analyzed in 2021. Results reveal that the proportion of high income, the proportion of rental income, and the number of apartments positively affect apartment prices, whereas the proportion of low income, bus stop, and the number of multi-family houses negatively affect apartment prices.