머신러닝 기반 한국 임야 공매의 낙찰가격 예측

This study explores factors influencing forest land bids in public auctions and predicts winning prices. Data from 8,394 forests sold via ONBID from 2016 to 2023 were analyzed, including property details, bid information, and economic indicators like population, taxes, and land price changes. PCA and MLR were used to identify key independent variables, and DT, RF, SVR, and XGB algorithms were employed to improve winning bid price accuracy. MAE and MAPE assessed prediction performance. Key factors increasing winning bid prices include the previous year’s land price change rate, number of inquiries, number of creditors requesting distribution, initial bid price, final bid price, and land size. Factors decreasing winning bid prices are whether a property is landlocked, whether the appraisal report contains the word “impossible,” the previous year’s local income tax in the area where the land is located, and whether the property is sold as partial ownership. Initially, the previous year’s land price change rate increased winning bid prices but decreased as the auction progressed. SVR demonstrated the highest prediction accuracy, followed by MLR, DT, RF, and XGB. Hyperparameter optimization enhanced predictive accuracy, particularly challenging for first-round bids. MAPE improved to 4.17% from the second to the sixth rounds. This study effectively predicts winning bid prices with high accuracy, would apply to auctions and real estate, and be valuable for general land price evaluations.