Utility Pole Damage Prediction for Heavy Rainfall by Machine Learning Technique with Public Data

Abstract

When heavy rain hits, some utility poles for telecommunication were damaged because of landslides or road collapses. These damages may cause telecommunication disruptions, and this paper aims to build damage prediction to make countermeasures for mitigation and restore early. In this paper, we aim to construct a prediction technique that can estimate the damage to individual utility poles by using the damage results of southern Kumamoto during heavy rainfall in July 2020. In addition, the prediction model uses public data that are maintained nationwide to enable prediction at any location. As a result, we built a highly accurate prediction model with a PR-AUC of 0.71 in the test data and confirmed that the model can be interpretable.