Support system for road price formulation using machine learning

Fixed asset tax is assessed based on land use and is an essential financial resource for Japanese municipalities. Therefore, the evaluation of fixed asset tax is important for local governments. A real estate appraiser uses the road price index to determine the fixed asset tax on land; therefore, the appropriate road price should be specified. The local government typically decides the road price based on expert experience and knowledge. Hence, there is need for an objective and transparent basis for road price decisions, and municipal officials currently devote considerable time and effort to surveys for appropriate evaluation. In this study, large-scale data, such as public and private sector data, with machine learning, are utilized to construct a road price estimation system under an industry?government?academia collaboration. An ensemble model utilizing a machine-learning algorithm was applied, i.e., the gradient boosting decision tree, to learn previous road price data and the various factors that affect road prices. In addition, via visualization, the importance of each element that affects road prices was quantified, thereby enabling the determination of the effect of each element or feature objectively. Our proposed system can be used to develop a road price estimation model and predict road prices for new roads. Its usage is aimed at promoting data utilization in local governments to reduce manual labor as well as to improve efficiency and transparency in formulating road prices. Furthermore, it creates new possibilities for the activation and utilization of public and private sector data, such as fixed asset data.