Improving Off-System Bridge Monitoring: An AI-Driven Methodology for Enhanced Condition Prediction

Abstract: Off-system bridges are critical components of the United States' transportation infrastructure, providing essential access to rural communities and enabling residents to reach vital services such as employment, education, and healthcare. However, many of these bridges are structurally deficient or functionally obsolete, and their deterioration is exacerbated by limited monitoring and maintenance. The closure of these bridges disproportionately hinders the mobility of underserved populations, worsening socioeconomic disparities—a concern highlighted by the Bipartisan Infrastructure Law (BIL). Despite existing research on bridge condition prediction utilizing probabilistic, or artificial intelligence frameworks, there is often insufficient focus on the unique challenges posed by off-system bridges. This includes evaluating condition ratings, and addressing class imbalances, largely due to the distinct nature of the modeling frameworks. This paper introduces an innovative method for predicting bridge deck conditions using data from the Texas Department of Transportation (TxDOT). The proposed approach leverages the strengths of probabilistic, and AI frameworks by integrating Generative Adversarial Networks with Focal Loss (GAN-FL) to generate synthetic data, thereby enhancing precision-recall balance in imbalanced datasets when combined with Random Forest models. The study finds that integrating GAN-FL into the Random Forest Classifier significantly enhances the performance of minority classes, improving their precision, recall, and F1 scores.