|
Title |
Future Damage Index Prediction of Steel Box-Girder Bridges Using SHAP-Based Machine Learning
|
|
Authors |
왕길환(Gil Hwan Wang) ; 부지연(Ji Yun Boo) ; 전종수(Jong-Su Jeon) |
|
DOI |
https://doi.org/10.11112/jksmi.2026.30.5.160 |
|
Keywords |
성능 예측; 강합성상형교; 교량 유지관리 자료; 머신러닝; SHAP Damage index; Steel box-girder bridge; Bridge maintenance data; Machine learning; Shapley additive explanations |
|
Abstract |
This study develops an interpretable machine-learning framework for predicting the future damage index of steel box-girder bridges using readily available maintenance data. A database of 1,066 inspection transition records from 194 bridges was constructed by integrating bridge specifications and inspection histories from the Facility Management System, meteorological observations from the Korea Meteorological Administration, and annual average daily traffic from the Traffic Monitoring System. Three boosting-based regression algorithms were compared after min-max scaling and Optuna-based hyperparameter optimization. Extreme gradient boosting achieved the best holdout performance, with a root mean square error of 0.020 and a coefficient of determination of 0.823. Shapley additive explanations analysis showed that the current damage index, inspection interval, and bridge age were the dominant predictors, whereas the freeze-thaw exposure, chloride-related exposure, and traffic volume had moderate effects. The proposed model can support rapid screening and maintenance prioritization when detailed inspections are limited.
|