What question did this study set out to answer?

The study aims to improve the accuracy of forecasting preventive maintenance costs for expressways using advanced machine learning techniques.

June 18, 2026Open Access

Machine learning approaches for predicting preventive maintenance costs of expressways in Xinjiang

Key Points

The study aims to improve the accuracy of forecasting preventive maintenance costs for expressways using advanced machine learning techniques.
Developed three hybrid models optimized with different algorithms for specific maintenance measures.
Used data from expressways in Xinjiang for model training and validation.
Conducted statistical tests to assess model performance and robustness.
FOA-XGBoost achieved R² = 0.8622 and MAE = 0.0552, improving XGBoost's performance.
HOA-RF outperformed RF with MSE = 0.0032 and RMSE = 0.0565.
PSO-BPNN achieved the highest R² of 0.9277 and the lowest MAE of 0.0419 among all models.

Abstract

With the expansion of high-speed road mileage and the aging of road networks, maintenance workloads have increased dramatically. Accurately forecasting preventive maintenance costs is a practical approach to rationally control maintenance expenditure, enhance capital utilization efficiency, and ensure the long-term stable operation of the road network. However, the current predictive model for preventive maintenance costs fails to adequately account for variations in types of preventive maintenance measures, limiting their ability to capture the actual cost variation patterns under different engineering conditions. Consequently, there remains scope for improvement in prediction accuracy. In this study, we developed three meta-heuristic-optimized hybrid models, each tailored to a specific asphalt pavement preventive maintenance measure: 1) the fruit fly optimization algorithm (FOA)-enhanced XGBoost model for crack filling; 2) the hiking optimization algorithm (HOA)-enhanced RF model for surface sealing; 3) the particle swarm optimization (PSO)-enhanced BPNN model for overlay. A case study demonstrates that FOA-XGBoost (R² = 0.8622, MAE = 0.0552) improved XGBoost’s R² by 0.0526. In contrast, HOA-RF (MSE = 0.0032, RMSE = 0.0565) outperformed RF with lower error metrics. Furthermore, PSO-BPNN achieved the highest R² (0.9277) and the lowest MAE (0.0419) compared to BPNN. All models maintained MAPE below 5%. To further support the reliability of the results, the Wilcoxon signed-rank test was conducted to assess the statistical significance of model performance differences. In addition, nested cross-validation and sensitivity analysis were performed to evaluate the robustness and stability of the proposed models. These findings indicate that the optimized hybrid model has demonstrated improvements in both predictive accuracy and stability. Consequently, the developed models may provide useful support for maintenance cost estimation and resource allocation in expressway maintenance management.

Read Full Paperexternally

Mark Helpful

Bookmark

Relay

View Full Paper