Comparative AI-Based Forecasting of Pavement Conditions under Maintenance and No-Maintenance Scenarios

Document Type

Conference Proceeding

Publication Date

1-1-2026

Abstract

Accurate forecasting of pavement conditions supports strategic infrastructure decisions and maintenance planning. Although the existing literature has examined the prediction of future pavement conditions under no-maintenance scenarios, there remains a critical need for models that also account for the effects of maintenance activities. Reliable prediction of post-maintenance pavement conditions is essential for selecting cost-effective treatments, forecasting life-cycle performance, and prioritizing limited maintenance budgets. Therefore, this study conducts a comparative analysis of International Roughness Index (IRI) condition predictions under both maintenance and no-maintenance scenarios using machine learning (ML) models. The models incorporate traffic, structural, and environmental variables as input features for 2-year and 3-year horizon predictions. Results showed that the no-maintenance scenario yielded higher prediction accuracies, whereas the maintenance scenario introduced some challenges. The XGBoost model achieved the most effective performance by R2, RMSE, MAE, and WMAPE and considering computational efficiencies. A performance comparison is conducted across different maintenance activities. For the 2-year horizon, IRI prediction under resurfacing yielded the highest accuracy, whereas the 3-year horizon proved most challenging to predict IRI under maintenance. Predictions under thin overlays showed almost similar accuracy across both horizons, whereas thick overlays were comparatively easier conditions to forecast IRI within the 2-year horizon.

Publication Title

Construction Research Congress 2026 Advanced Technologies Artificial Intelligence and Data Analytics in Construction Selected Papers from Construction Research Congress 2026

ISBN

[9780784486962]

Link to Full Text

Share

COinS