랜덤 포레스트를 활용한 작품 가격 예측 모형 연구

Abstract

Purpose: Despite the growing art market, few studies have been conducted on the model for estimating the price of artworks. The objective of this paper is to develop a model to estimate art price, and to compare and analyze the predictive performance of nonparametric models. Methods: This study used nonparametric models (e.g., random forests, k-nearest neighbor, support vector regression) to predict the price of artworks and compared their prediction accuracy performances based on root mean square error (RMSE). An evaluation was carried out on five years of Korean auction data from 2014 to 2018. The performance of the model can be improved by selecting appropriate hyperparameters and be generalized through 10-fold cross-validation. Results: According to the results of comparing the predicted performance based on the RMSE calculated for each fold optimization model, the random forest model predicted the best performance. In particular, the prediction error of the random forest model was about 40% lower than that of the parametric OLS model. Conclusion: This study proves the applicability of nonparametric models to estimate art prices empirically. The developed model described in this paper reduces the transaction cost of artworks and reduces the limitations of the art appraisal system.