A deep learning algorithm-driven approach to predicting repair costs associated with natural disaster indicators: The case of accommodation facilities

Abstract

It is well known that accurate and reliable maintenance and repair cost estimates are important to maintain a building in its optimal condition, especially during the operation and maintenance phase within the whole life cycle. However, due to emerging trends in buildings that are high-performance, large-scale, complex, and high-rise, it is difficult to achieve those cost estimates. In addition, the impact of climate changes that tend to occur more frequent and severe natural disasters has caused increasing damages to buildings , yet little is still specifically known about predicting the impact of natural disasters on repair costs of accommodation facilities accurately and reliably. This study fills this gap by developing and validating a deep neural network (DNN) model that can generalize repair cost trends associated with natural disaster factors, including peak ground acceleration , precipitation, wind speed, geographic profiles of adjacent water systems, drawing on 1125 insurance claim payout records on accommodation facilities. The robustness of the developed DNN model was scientifically tested and validated using the root mean squared error and the mean absolute error methods. Practical applicability of the proposed modeling framework was then demonstrated by creating predicted repair cost trends. This study contributes to the existing knowledge by proposing a deep learning method that predicts repair costs of accommodation facilities associated with natural disasters, while providing both facility managers and insurance companies with evidence-based reference to develop better-informed cost management strategies against potential natural disasters. • Prediction of repair costs of accommodation facilities against natural hazards. • Development of a deep neural network (DNN)-driven learning model. • Improved predictability of the DNN model compared to the multiple regression model. • Model validation using mean absolute error and root mean squared error methods. • Repair cost trends by weather and seismic conditions and water system profiles.