Use of artificial intelligence to predict the accuracy of pre-tender building cost estimate

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