Estimate accuracy (bias) was used as the output variable. The prediction power stands at 73% correlation coefficient, 3% of Mean Absolute Error and 0.2% Mean Squared Error. It was found that in more than 73% of the test cases the predicted estimate bias did not differ by more than 8.2% from the expected (Maximum Absolute Error). This means that amount of estimate bias predicted by the ANN are similar to what actually occurred. The trained ANN model can be used as a decision making tool by cost advisors when forecasting building cost at the pretender stage. The model can be queried with the characteristics of a new project in order to quickly predict the error in the estimate of the new project. The predicted error represents the additional contingency reserve that must be set aside for the project in order to cater for possible cost overruns. The model can also be extended to forecast the likely cost of a project.Pre-tender estimates are susceptible to inaccuracies (biases) because they are often prepared within a limited timeframe, and with limited information about project scope. Inaccurate estimation of project uncertainties is the underlying cause of project cost overruns in construction. Typically, cost engineers and quantity surveyors would add contingency reserve to a pretender estimate in order to account for any unforeseen cost that may arise between the date of the estimate and the projected completion date of the project. The traditional 10% rule of thumb for estimating contingency is subjective - based on experience and expert judgment, and are often inadequate. In the research reported in this paper, we propose that learning algorithms trained to use the known characteristic of completed projects could allow quantitative and objective estimation of the inaccuracies in pretender building cost estimates of new projects.The study assumes that the accuracy in the initial estimate (bias) of a completed project is the difference between the actual project completion costs minus the pre-tender cost forecast expressed as a percentage of the actual project completion costs. A three- layer ANN model of feed- forward type with one output node was constructed and trained to generalise nine characteristics of 100 completed projects and the cost data from those projects. The nine input variables of the model are project size (measured by number of storeys and gross floor area), principal structural material, procurement route, project type, location, sector, estimating method, and estimated sum.
|Title of host publication||MISBE2011|
|Subtitle of host publication||Management and Innovation for a Sustainable Built Environment|
|Place of Publication||Delft, The Netherlands|
|Publisher||Delft University of Technology|
|Number of pages||1|
|Publication status||Published - 2011|
|Event||International Conference on Management and Innovation for a Sustainable Built Environment (MISBE) - Amsterdam, The Netherlands, Netherlands|
Duration: 19 Jun 2011 → 23 Jun 2011
|Conference||International Conference on Management and Innovation for a Sustainable Built Environment (MISBE)|
|Period||19/06/11 → 23/06/11|
Aibinu, A., Dassanayake, D., & Chau Thien, V. (2011). Use of artificial intelligence to predict the accuracy of pre-tender building cost estimate. In MISBE2011: Management and Innovation for a Sustainable Built Environment (pp. 17). Delft University of Technology.