Extraction of Road Maintenance Criteria using Machine Learning and Spatial Information

Infrastructure maintenance requires extensive financial and human resources, and a lack of these resources---and, in particular, a shortage of experts---is a problem in many countries and regions around the world. In response to such circumstances, there is considerable research on infrastructure damage-detection methods using camera images and machine-learning. However, even if a large number of damaged parts are found using such methods, the decision whether to repair damaged areas is nevertheless determined empirically, by taking into account several factors such as road statistics and the regional characteristics. For these reasons, the current situation is that municipalities that lack experts cannot make comprehensive decisions regarding repairs. Therefore, in this research, we extracted maintenance management standards and automated decision-making using the decisions made by local government officials regarding damaged roads in Japan. We focused on roads, because roads are considered to be one of the most influential infrastructure. In order to do so, we cooperated with six municipalities in Japan. We combined statistical information regarding damaged roads with regional characteristics. As a result, in a very understandable way, we were then able to reproduce the decisions made by experts with an accuracy of 0.75. Our research has the potential to enable automated decision-making in the future.