Abstract
Both in the Tokyo metropolitan area and rural areas in Japan, the number of development projects being undertaken by private companies and local governments has been increasing. Under these circumstances, there is need for a system to estimate the rent paid by tenants and consider the features of living areas from several perspectives. This information can be used by developers when choosing a site. For both price estimation and an analysis of factors affecting price, we apply three approaches: deep neural networks, the hedonic approach, and random forest regression and clarify the advantages and limitations of them. The results show the importance of data cleaning when using deep neural network. Also, they show that it is possible to create a potential map of areas and visualize time series transformations and the impacts of economic trends on it by using latitude, longitude, the year of registration and deflator as explanatory variables besides building attributes.
