Citation

Kong X, Li Z, Zhang Y, Das S. Transp. Res. Rec. 2022; 2676(7): 570-584.

Copyright

(Copyright © 2022, Transportation Research Board, National Research Council, National Academy of Sciences USA, Publisher SAGE Publishing)

DOI

10.1177/03611981221080140

PMID

unavailable

Abstract

The deck condition of bridges is one of the most important factors impacting the connectivity and efficiency of transportation networks. Bridges with quickly deteriorating deck conditions are a huge financial burden for transportation agencies and can downgrade the efficiency of the whole transportation network. This study utilizes an interpretable machine learning framework, Shapley additive explanation (SHAP), to investigate the associations between various factors, such as wearing surface, deck structure, and so forth, and bridges with quickly deteriorating deck conditions nationwide. An XGBoost model is trained to perform the binary classification task on a heavily imbalanced dataset and classify relatively young bridges (less than 20 years old) with poor/fair deck conditions and relatively old bridges (30-40 years old) with good deck conditions in the National Bridge Inventory (NBI) database. The accuracy of the predictive model is 0.91, and the AUC score is about 0.83. After applying this well-performed predictive model on the interpretable machine learning framework, the results revealed that without wearing surface, corrugated steel deck structure, wide bridge structure, and long span are highly associated with bridges with quickly deteriorating decks. The results also show that bridges with a relatively low average daily traffic (ADT) or truck percentage of ADT are in a dilemma zone, where the overall traffic or truck volume of the bridge is not low enough to prevent fast deterioration, but not high enough for eligibility for the funding required for more durable materials during construction or appropriate maintenance.

Language: en

Keywords

artificial intelligence; artificial intelligence and advanced computing applications; bridge and structures management; bridge condition data/assessment; bridge data QC/QA; bridges and structures; construction; data analytics; data and data science; infrastructure; infrastructure management and system preservation; machine learning (artificial intelligence); pattern recognition