Multistage Cost Estimation Framework for Transparent and Granular Bridge Maintenance

Abstract Bridge maintenance cost is not an isolated variable but a constructed outcome derived from damage size, repair method, and unit cost. However, existing data-driven approaches often treat cost estimation as a black-box prediction problem, providing only aggregated cost values, without transparent and granular information. To address this knowledge gap, this study proposes a multistage cost estimation framework that explicitly structures maintenance cost estimation according to the fundamental construction management logic of quantity–method–price. The proposed framework sequentially estimates repair-required damage size, classifies appropriate repair methods, determines corresponding unit costs, and integrates these components to total maintenance cost. An extreme gradient boosting–based pipeline was developed using 3,716 inspection records from the Korean Bridge Management System and validated with 100 real-world bridge maintenance records. The results show that the proposed framework reduces the mean absolute error by 41.3% and 73.7% compared to conventional direct cost estimation models. By aligning data-driven modeling with the maintenance cost formation logic, this study advances bridge maintenance with more transparent and granular cost estimation.