Keywords: bidding, classification, cost, data mining, highway.

Construction project costs often increase above the original bid price. Early information about a project likely to have large cost overruns could be useful to the owners to plan for possible increased costs and to more closely monitor a project that has the potential for large overruns. Data available at the time of the bid opening can possibly be used to produce a prediction or an indicator of projects with the potential for large cost overruns.

Data for this analysis was collected at random from the website of the California Department of Transportation. All of the projects were bid in 2006. Data from 275 competitively bid highway projects were collected. The average project was completed at a price 2.91% greater than the original bid amount. The project data collected varied widely in cost magnitude and type of construction. Some projects were maintenance projects while others were major rehabilitations or new construction.

The data was initially examined using the treemap data visualization technique [1]. This analysis suggested that projects that have costs concentrated in a few project line items tend to have smaller cost increases during construction than projects with costs spread evenly over many items.

The K* classification algorithm [2] was employed to predict project cost overruns. The Weka data mining software [3] was used to implement the K* algorithm. Inputs to the model included a measure of the bid dissonance, the cost overrun for the project and the percentage of total project cost of the two largest project line items. A prediction of the percentage cost increase of the project for a test set was the output. Both the total data set and a data set with outliers removed were analyzed.

Analysis of the output shows that the total data set provides better predictions for large deviations between the low bid and completed project cost. The trimmed data set however provided more accurate predictions. Cases where the project cost was significantly lower than the bid amount were not predicted successfully by K* with either data set. This application of the K* algorithm suggests that there may be predictable patterns in the bidding data that suggest that when a project will have a large cost increase during construction. It is suggested that the K* model could be used as an indicator of the potential for large cost increases.

1

B.B. Bederson, B. Schneiderman, M. Wattenberg, "Ordered and Quantum Treemaps: Making Effective Use of 2D Space to Display Hierarchies", ACM Transactions on Graphics, 21(4), 21, 2002. doi:10.1145/571647.571649

2

J.G. Cleary, L.E. Trigg. "K*: an instance-based learner using an entropic distance measure", in "Proceedings of 12th International Conference on Machine Learning, 9-12 July 1995", Morgan Kaufmann Publishers, San Francisco, CA, USA, 1995.

3

M. Hall, E. Frank, G. Holmes, B. Pharinger, P. Reutmann, I.H. Witten, "The Weka Data Mining Software: An Update", SIGKDD Explorations, 11(1), 2009.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £130 +P&P)