Artificial neural networks for construction bid decisions

Abstract

An Artificial Neural Network (ANN) approach was explored for supporting construction bid decisions, since such decisions are heavily dependent on practitioner expertise, which in turn is generally encapsulated in case histories. One of the ANNs described here was trained on knowledge from a sample of the entire Sri Lankan construction industry, and was used to predict the preferred job sizes for firms of differing characteristics; such information could help firms in their bid/no-bid decisions. The other ANN was trained on case histories elicited from a single contractor, and was used to predict the percentage mark-up. The network outputs were obtained in both binary output and continuous valued output formats. The former format had some distinct advantages over the latter, as it provided greater information for decision making instead of being a "black box" output. The influences of the middle layer size, output format and allowable error during training, on the training duration and accuracy of prediction were studied.