Machine-Learning Models for Predicting Condition Rating of Cast-in-Place and Precast Concrete Bridge Decks

Abstract This study presents a comparative evaluation of six machine learning models for predicting the condition ratings of cast-in-place and precast concrete bridge decks using a large-scale nationwide bridge data set. The study combines machine learning techniques with genetic algorithm-based hyperparameter optimization to improve prediction performance and support bridge maintenance and rehabilitation decision-making. The models were developed in four stages: data collection of bridge inspection and characteristics records for 286,034 bridges across all 50 US states from the National Bridge Inventory database; data preprocessing to clean, transform, and split the collected data set into training and testing sets; model training using ordinary least squares, ridge regression, least absolute shrinkage and selection operator regression, support vector regressor, random forest, and extreme gradient boosting; and model evaluation using the five performance metrics of coefficient of determination, mean absolute percentage error, mean absolute error, median absolute error, and root mean square error, in addition to fivefold cross-validation, a testing data set, and permutation importance analysis. The results show that the extreme gradient boosting and random forest models outperformed the other developed models across all five performance metrics. The developed models enable transportation agencies to predict the condition ratings of cast-in-place and precast concrete bridge decks across the United States and accordingly prioritize their maintenance and rehabilitation decisions.