A Machine Learning Framework for Early-Stage Cost Estimation of Transportation Projects

To address the limitations of traditional planning-level cost estimating methods, such as reliance on statewide averages, heavy reliance on engineering judgment, and high levels of inaccuracy, this study developed two data-driven frameworks and corresponding tools. The first framework focused on resurfacing projects and used treatment-specific simple linear regression based on lane miles and geographic location. The second addressed a broader group of construction projects and used multiple linear regression with project length, route type, and right-of-way cost as key predictors. In both frameworks, the final estimates are computed by aggregating multiple estimates produced by county-, region-, and state-level models. The resurfacing framework produced more stable results, with an overall MAPE of 32.45% and MPE of -1%. The construction framework showed weaker performance, with an overall MAPE of 143.91% and MPE of 106.04%. Both frameworks produced estimates that generally align with industry standards for the expected accuracy of early-stage cost estimates.