Abstract

Accurate estimation of highway maintenance costs is essential for efficient resource allocation and long-term infrastructure sustainability. This study evaluates the effectiveness of advanced machine learning models, including ResNet, Transformer, and other neural network architectures, for forecasting maintenance costs using historical data from the Highway Maintenance Improvement Program (HMIP) provided by the North Carolina Department of Transportation (NC DOT). A comprehensive comparative analysis is conducted across multiple models using standard performance metrics, including R2, MAE, RMSE, and MSE. The results demonstrate that advanced architectures, particularly ResNet and Transformer, consistently outperform traditional statistical approaches and baseline machine learning models, achieving R2 values exceeding 0.95 and substantially lower prediction errors. In addition, an ablation study is performed to assess model robustness under reduced feature availability, showing that the proposed models maintain strong predictive performance even with a limited set of key variables. These findings highlight the ability of advanced machine learning models to capture complex nonlinear relationships in structured infrastructure data and demonstrate their practical value for improving decision-making in maintenance planning and life-cycle cost analysis (LCCA).

Department(s)

Engineering Management and Systems Engineering

Publication Status

Open Access

Comments

University Transportation Centers, Grant 69A3552348339

International Standard Serial Number (ISSN)

1547-2701; 0013-791X

Document Type

Article - Journal

Document Version

Citation

File Type

text

Language(s)

English

Rights

© 2026 Taylor and Francis Group; Taylor and Francis, All rights reserved.

Creative Commons Licensing

Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.

Publication Date

01 Jan 2026

Download

Full Text Link

Share

 
COinS