Improving the Accuracy of Early Cost Estimates on Transportation Infrastructure Projects

Abstract

A better understanding of top-down estimating practices and their contribution to budgeting accuracy allows public transportation agencies to allocate limited construction funds more efficiently. This paper builds on a recent study that evaluated the accuracy of early highway construction cost estimates for the Montana Department of Transportation (MDT). The study included 996 MDT projects awarded between 2006 and 2015, with more than $2.2 billion in construction costs, accounting for more than 82% of the agency’s construction spending. The results suggest that top-down models provide a means to improve the prediction accuracy of agency cost estimates (when measured as the mean absolute percentage error of project costs), particularly for projects with higher levels of complexity and lower sample sizes. These conclusions are drawn from a comparison of agency in-house estimates to predictions obtained through artificial neural network (ANN) and multiple regression models. In interpreting these findings, the paper demonstrates that the bias-variance trade-off, a common model building concern in the machine learning and artificial neural network literature, is likely a key factor in explaining the prediction performance of simplified models.