Predicting Ratio of Low Bid to Owner’s Estimate Using Feedforward Neural Networks for Highway Construction

Abstract

State Departments of Transportation (state DOTs) are constantly challenged by accurate estimate cost of highway projects. The significant deviation between the lowest submitted bids and the owner’s estimate is problematic for state DOTs that strive to deliver projects on time and within budget. However, few empirical studies focused on narrowing the gap between low bids and owner’s estimate. The major objective of this research is to enhance the quality of owner’s estimate to more accurately forecast the lowest submitted bid. To achieve this objective, the feedforward neural network in the framework of Friedman’s model is utilized to enhance owner’s estimate. A pool of potential variables to explain the cost variability are selected as input attributes, representing bidding environment, construction market, macro-economic market, and energy market conditions. Bid tabulations for highway construction projects let by the Georgia Department of Transportation (GDOT) from 2010 to 2018 are utilized to conduct empirical study and quantitatively measure the model performance. The outcome shows that the proposed model significantly reduces the deviation between owner’s estimate and the submitted low bid. This study contributes to the body of knowledge through creation of an artificial neural network model that improves the accuracy of owner’s estimate and reduces the gap between owner’s estimate and the lowest submitted bid. It is anticipated that transportation professionals benefit from the results of this study to prepare more accurate cost estimates and budget plans for enhanced decision-making.