Predicting Pavement Condition Index for Cold‐Region Highways Using an ENET‐XGBoost Coupled Model

ABSTRACT

Highway pavement in cold regions endures long‐term deterioration caused by the combined effects of traffic loads and environmental conditions. This degradation follows a complex mechanism that traditional linear models struggle to represent accurately, as they often overlook nonlinear interactions and multicollinearity among variables. To address this challenge, we propose a two‐stage prediction framework combining elastic net (ENET) and XGBoost. Initially, ENET extracts underlying trends and selects relevant features by applying mixed regularization, which filters out less influential factors from the Pavement condition index (PCI) analysis. Subsequently, XGBoost models the residuals left by ENET, capturing intricate nonlinear relationships among the variables. Our case study employs traffic and meteorological data collected on a highway in Inner Mongolia between 2020 and 2025. The ENET stage excludes precipitation and Sunshine Days due to their weak correlation with PCI, while retaining cumulative equivalent axle passes, average annual daily traffic (AADT), and extreme minimum temperature as primary predictors. The nonlinear residual correction by XGBoost reveals the complex coupling between load and temperature effects. The combined model achieves an R ² of 0.969 in full back‐calculation, with a mean relative error of 0.70%. Under leave‐one‐out cross‐validation, the coupled model yields an R ² of 0.699, providing a conservative estimate of generalization performance. Markov state transition analysis indicates that the current PCI grade is classified as Good, with a 75.0% chance of remaining in this state through 2026. Predictions across multiple scenarios suggest that traffic load factors exert a stronger influence on PCI degradation than climatic factors. Drawing on these findings, we propose staged preventive maintenance strategies, offering a data‐driven foundation for more targeted highway upkeep.