1. Home
  2. TRID
  3. View Record
https://trb-annual-meeting.nationalacademies.org/registration

Application of ANN Model to Determine Primary Drivers of Road Construction Cost

Increasing cost as well as cost overrun in road projects is a worldwide reality, and efforts to determine cost drivers have therefore been a primary exercise undertaken by the stakeholders and even those with academic interests. Identifying the primary cost drivers becomes important because models to predict unit cost of road construction projects are generally pivoted around them. Given the established advantages of ANN models in dealing with noisy data as well as nonlinear relationships between the variables, this paper outlines an approach based on the relative importance of each variable, which can be used to rank the variables and decide on the primary cost drivers for road construction. Such relative importance is determined from information stored in the connection-weights of an ANN model, employing Garson’s algorithm. In addition, the paper shows that the principal component analysis (PCA) gives further insights, thereby complementing the determined relative importance. Now, one of the main reasons that the stakeholders are interested in knowing the primary drivers is that, once known, then the focus can shift to factors which dictate the cost for each individual cost driver. Given that variations in cost (for any particular cost driver) are contingent upon three factors, namely, bid-quantity, bidded-rates, and project-length, it becomes imperative that questions are asked on how much they influence the cost (variations in the cost data) and also their relative influence. The paper demonstrates that a similar relative importance approach using connection-weights of an ANN model provides a means to flag the extents to which each of the aforesaid three factors impacts variations witnessed in the cost of any individual cost-driver. The results further demonstrate that the length of a project is indeed an important parameter, which lends a scale-effect when it comes to unit cost (cost per unit length), corroborating the widely perceived view that smaller projects tend to have a larger unit cost.

Language

  • English

Media Info

  • Media Type: Web
  • Features: References;
  • Pagination: pp 106-118
  • Monograph Title: International Conference on Transportation and Development 2023: Transportation Planning, Operations, and Transit

Subject/Index Terms

Filing Info

  • Accession Number: 01886897
  • Record Type: Publication
  • ISBN: 9780784484883
  • Files: TRIS, ASCE
  • Created Date: Jun 30 2023 11:28AM