Data-Driven Bridge Deck Condition Rating Prediction Using Machine Learning Models

Bridges are integral components of transportation networks, facilitating connections and supporting traffic loads across various obstacles. While bridges are important, they are costly to maintain because of high usage and strict inspection demands. With approximately 57,000 bridges in Texas, the state requires an accurate and computationally efficient deterioration model to support informed rehabilitation and reconstruction decisions by bridge management agencies. The deteriorating condition of bridges, along with escalating maintenance and rehabilitation costs, necessitate difficult budgeting choices to ensure that bridges remain safe with a prolonged operational lifespan. Machine learning (ML) techniques have become increasingly popular for developing predictive models and assisting infrastructure management decisions. Although these approaches show strong potential, their practical application is often limited by issues such as reliance on inspection-based ratings that may carry inherent biases ( 1 , 2 ) or the use of simplified or narrow classification categories ( 3 , 4 ). This study addresses these limitations by selecting exclusively inventory-based features without inspection biases, using a multi-class classification approach, and thoroughly optimizing multiple ML models. Seven different ML models, including Decision Tree (DT), K-Nearest Neighbors (KNN), Support Vector Machines (SVM), Random Forest (RF), Adaboost (AB), CatBoost (CB), and Bagged Tree (BT), were applied to the bridges within Texas. The prediction accuracy of the best model was found to be above 90%. Among the selected input features, bridge age proved to be the most significant factor affecting deck Condition Rating (CR), followed by span type and number of spans. Climate zones also showed substantial correlation.