Federal Highway Administration State Planning and Research Funding
Montana Department of Transportation
Hyung Seok "David" Jeong
About the research
This project contains the information and background on top-down cost estimating using artificial neural networks (ANN) to enhance the accuracy of MDT early estimates of construction costs. Upon conducting an extensive review of MDT’s budgeting and cost estimating efforts, and following a survey of agency experts on the identification of the most salient project attributes with the dual-objectives of low effort and high accuracy, a rational method for top-down variable selection is proposed.
Selected variables were further tested in their explanatory power of construction costs through the application of two cost estimating methodologies—multiple regression and artificial neural network methodologies. Both methods are shown to provide sizeable improvements over the agency’s current levels of prediction accuracy for its construction costs. Potential accuracy gains are also demonstrated to depend on project work types. The comparison of mean absolute percentage errors across different estimating methods confirms that the potential benefits from the proposed methodologies are expected to rise as the project level complexity and uncertainty increase. New construction and bridge replacement projects, for instance, are expected to gain the most in estimating accuracy since these two groups seem to exhibit considerably higher levels of deviation from the MDT’s preliminary cost estimates.
To facilitate MDT’s implementation of the suggested methodology described in this report, a cost estimation methodology was also presented in an Excel spreadsheet format. This achieves two goals. First, it provides an accessible tool to make top-down cost predictions for agency planners during the budgeting stage based on MDT’s historical project data. Second, it furnishes a process through which the proposed model can be improved as new project information becomes available. Ultimately, the insights gained from this study are expected to contribute to a better formulation of the agency’s early cost estimation and budgeting efforts.