Developing Predictive Models of Superstructure Ratings for Steel and Prestressed Concrete Bridges

A large number of deficient bridges may endanger the public and affect the economy at a broader scale. Bridge superstructure rating is a critical element that affects the overall sufficiency rating of a bridge. Accurately predicting the superstructure performance of a bridge may help agencies better prioritize their resources for maintenance and repairs. The main objective of the paper is to utilize data mining techniques to develop reliable models to predict the superstructure rating of bridges. This research utilizes the national bridge inventory (NBI) database as the main source of information. A focused subset was created based on the defined scope of the research: year built (≥ 1955), kind of material-design (prestressed concrete and steel), type of design (stringer/multi-beam or girder), and deck type (concrete cast-in-place). This paper takes three approaches for model development including linear regression, decision tree, and neural network. The best model was identified for each superstructure material through comparisons among different models. In addition, a discussion of individual variables and their contributions to predict superstructure rating was performed. The identified models provide insight into when a bridge superstructure needs maintenance and reconstruction.