Dealing with construction cost overruns using data mining

One of the main aims of any construction client is to procure a project within the limits of a predefined budget.
However, most construction projects routinely overrun their cost estimates. Existing theories on construction
cost overrun suggest a number of causes ranging from technical difficulties, optimism bias, managerial incompetence
and strategic misrepresentation. However, much of the budgetary decision-making process in the early
stages of a project is carried out in an environment of high uncertainty with little available information for accurate
estimation. Using non-parametric bootstrapping and ensemble modelling in artificial neural networks, final
project cost-forecasting models were developed with 1600 completed projects. This helped to extract information
embedded in data on completed construction projects, in an attempt to address the problem of the dearth of
information in the early stages of a project. It was found that 92% of the 100 validation predictions were within
±10% of the actual final cost of the project while 77% were within ±5% of actual final cost. This indicates the
model’s ability to generalize satisfactorily when validated with new data. The models are being deployed within
the operations of the industry partner involved in this research to help increase the reliability and accuracy of
initial cost estimates.