Development of big data-driven approach for predicting bridge component deterioration: Case study in Taiwan

• Proposed a novel integrated framework that combines data-driven analytics with targeted bridge component maintenance strategies. • Developed an RF model with 88% accuracy for predicting bridge component deterioration. • Identified critical components needing urgent maintenance (D = 3). • Mapped moderate deterioration components (D=2) for short-term rehabilitation. • Demonstrate RF’s superiority over KNN (74.81%), SVM (78%), and Logistic Regression (72.3%) in accuracy, precision, recall, and F1-score. Aging, seismic activities, and extreme climate conditions increasingly impact key monitored national highway bridges, leading to severe deterioration and structural damage. Enhancing inspection and maintenance procedures for optimal resource allocation is a key priority for management authorities. This study analyzes 27,237 historical inspection records and employs a Random Forest (RF) model to predict deteriorating bridge component locations. The model identifies critical components requiring priority maintenance, including anti-fall devices, abutment ends, drainage facilities, and wing walls. The RF model achieves 88% prediction accuracy, outperforming K-Nearest Neighbors (74.81%),Support Vector Machine (78%) and Logistic Regression (72.3%). These results validate the RF approach for bridge deterioration classification, enabling data-driven maintenance prioritization and reducing inspection time.