•  
  •  
 

MACHINE-LEARNING ENHANCEMENT OF HIGHWAY PAVEMENT CONDITION INDEX PREDICTION AND LONG-TERM DETERIORATION FORECASTING

Abstract

Accurate forecasting of pavement condition is essential for performance-based asset management of roadways, yet existing deterioration models remain limited by their reliance on fixed curves or empirical regression relationships that do not adequately reflect site-specific conditions or the impact of maintenance records. This study addresses these gaps by developing a data-driven machine learning framework for the Highway Pavement Condition Index (HPCI) using annual HPCI datasets from 2010 to 2023. The methodology integrates two components: (1) comparisons between measured and predicted HPCI to evaluate model's accuracy and stability, and (2) long-term forecasts of pavement deterioration across the national expressway networks. Four base models—Linear Regression, Random Forest, Gradient Boosting, and Extreme Gradient Boosting—were evaluated alongside ensemble Voting and Stacking Regressors. Residual analyses confirmed statistical unbiasedness across models, while Durbin–Watson tests indicated no significant autocorrelation. Feature-importance assessments consistently highlighted initial HPCI, materials, age as dominant predictors. The Stacking Regressor produced the highest accuracy in both training and testing phases and was therefore adopted as the final model. As a result, the observed deterioration–recovery dynamics in HPCI were reproduced with high fidelity. Also, the proposed model supports long-term strategic planning by identifying future maintenance needs and modeling network-level performance trajectories, demonstrating clear potential for improving highway pavement management operations.

Acknowledgements

Korea Expressway Corporation and Korea

This document is currently not available here.

Share

COinS