Key points are not available for this paper at this time.
Effective prediction of bridge deck condition, here defined as the National Bridge Inventory (NBI) condition rating based on visual inspections, is essential for prioritizing maintenance and ensuring infrastructure safety. This study proposes an ensemble framework that integrates six machine learning models and ranks them using the MOORA multi-criteria decision-making method with entropy-based weighting. Using National Bridge Inventory data from South Carolina (2019–2024), Random Forest consistently outperformed other models, demonstrating superior accuracy and robustness. Sensitivity analysis further validated the reliability of the ensemble-based ranking approach. Among the six models evaluated, Random Forest achieved the highest predictive accuracy (R² = 0.9853, RMSE = 0.0353), followed by Gradient Boosting and ANN. The ensemble-based MOORA ranking and sensitivity analysis confirmed that Random Forest was the most robust and reliable model for predicting bridge deck condition. The findings highlight the potential of combining machine learning and multi-criteria decision-making to enhance predictive maintenance and data-driven bridge management strategies.
Afriyie et al. (Thu,) studied this question.