Keywords: bidding, classification, construction, data mining.

Data mining is the process of automatically discovering useful information in large data repositories. In this paper the focus is on using classification, a data mining technique to identify patterns in highway project bidding data that may provide a prediction of the magnitude of a projects cost overrun during construction. Potentially, a projects potential for cost overruns is related to the pattern of the submitted bids.

Williams [1] has calculated five ratios that describe the nature of the submitted bids. These ratios are a way of representing the relationships between bids for a project that are dimensionless and are not dependent on the project magnitude. The ratios include the second lowest bid ratio, the mean bid ratio, the maximum bid ratio and the coefficient of variation of the submitted bids. Projects completed near the original low bid amount tend to have lower values of the ratios. It was also noted that the elevated ratio values seem to occur for projects that have large cost increases.

The data used for this analysis are from highway construction and rehabilitation projects conducted by the Texas Department of Transportation from 1995 to 2000. Each record in the database consisted of the data for one project. Each project had a value defined for the five ratios described above, and for the magnitude of the low bid amount. These could take on the nominal values of "low", "medium", or "high". Each project also had its cost overrun recorded as "Near", "Overrun", or "BigOverrun". An "Overrun" project was completed for a cost overrun of between 5% and 10%. A "BigOverrun" project was one in which the completed project cost exceeded the low bid amount by more than 10%.

The Ridor (Ripple Down Rules) classification algorithm was applied to the bidding ratio data [3]. Six classification rules were generated. The generated rules produced correct predictions for 43.89% of the projects. However, the model correctly classified 100 out of 190 projects (53%) with large cost overruns. It was found to be difficult to classify the "Overrun" category because it has a smaller number of cases than the "Near" or "BigOverrun" categories. This exploration of the use of the Ridor classification algorithm showed some potential for producing rules that can predict construction project cost overruns. In particular, the rules are able to predict projects with a greater than 10% cost overrun with a success rate of 53%. However, examination of the bidding data indicates that the existence of some contradictory cases in the test set make it difficult for the Ridor algorithm to produce rules that are highly accurate. Additional research is required to find improved algorithms and data inputs that can better handle the irregularities in the data.

1

T. Williams, "Bidding Ratios to Predict Highway Project Costs", Engineering, Construction and Architectural Management 12, 38-51. doi:10.1108/09699980510576880

2

T. Williams, S. Lakshminarayanan, H. Sackrowitz, "Analyzing Bidding Statistics to Predict Completed Project Cost", in Proceedings of the ASCE Conference on Computing in Civil Engineering, CD-ROM, 2005. doi:10.1061/40794(179)157

3

I.H. Witten, E. Frank, "Data mining: practical machine learning tools and techniques", Morgan Kaufman Publishers, San Francisco, USA, 2005.

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 £62 +P&P)