Understanding construction bidding behaviour using data visualization and data mining

This paper describes the use of the K ^* classification algorithm to predict cost overruns on competitively bid highway projects. Data were obtained from the California Department of Transportation about competitively bid highway construction projects. This data included the dollar amount of all submitted bids, the final construction cost, and the percentage of total project cost for the two largest project line items. This data was initially analyzed using the treemap visualization technique. It was found that projects where the costs were concentrated in the two largest items were more likely to have lower levels of project cost overruns. Then the K ^* algorithm, an instance-based classifier, was used to make predictions of final project costs. Both the total data set and a data set with outliers trimmed were applied to the model. The predictions were compared with the actual outcome for the test set projects. The K ^* algorithm had some success in predicting projects that would have large cost increases.