2021 Volume 77 Issue 2 Pages 62-71
With the increase number in vacancies, researchers attempt to estimate the spatial distribution of vacant housing. In specifying the model, municipality-owned data are recently included due to the existence of features that contribute to the accuracy of the model. This study constructs a predictive model of future vacant housing distribution using municipality-owned data. We employ XGBoost, a machine learning method, to deal with the existence of missing values and non-linearity of data structure. In consequence, we obtain the following findings. First, since the model proposed is based on decision trees, which enable us to flexibly deal with complicated and missing data. Second, the important features we picked up follow the arguments by previous studies, and thus confirmed the validity of the proposed model. Third, the proposed model approximates 84.3 percent of accuracy per 125-meter grid cell, which holds the sufficient level of the accuracy for municipalities to take advantage of.
