Journal of Information Technology in Construction
ITcon Vol. 19, pg. 462-473, http://www.itcon.org/2014/27
Comparative study in the use of neural networks for order of magnitude cost estimating in construction
submitted: | January 2014 | |
revised: | September 2014 | |
published: | October 2014 | |
editor(s): | Rezgui Y | |
authors: | Hany El-Sawah, PhD, PMP, Professor Faculty of Engineering, Helwan University, Cairo Egypt; Visiting Scientist, Department of Building Civil & Environmental Engineering, Concordia University, Montreal, Canada; email: helsawah@hotmail.com Osama Moselhi, PhD, P.Eng., Professor Department of Building Civil & Environmental Engineering, Concordia University, Montreal, Canada; email: Moselhi@encs.concordia.ca | |
summary: | This paper presents a study on the use of artificial neural networks (ANNs) in preliminary cost estimating. The choice and the design of the ANN model significantly affect the results obtained from the model and, hence, the accuracy of the estimated cost. The study considered Back Propagation Neural Network (BPNN), Probabilistic Neural Network (PNN) and Generalized Regression Network (GRNN) as well as regression analysis. Models were developed for order of magnitude cost estimating of low-rise structural steel buildings and short-span timber bridges. The study was conducted on actual data for 35 low-rise structural steel buildings and their respective cost was estimated using the developed regression and ANN models. These models were also applied to estimate the cost of a timber bridge extracted from the literature. The results showed that the mean absolute percentage error (MAPE) for the neural network models ranges from 16.83% to 19.35% whereas was equal to 23.72% for the regression model. Moreover, the linear regression model was more sensitive to the change of the number of the training data and that the PNN network was the most stable network among all the other estimating models as the maximum difference in MAPE percentage was only 2.46%. Whereas, the maximum difference in MAPE was 19.47%, 17.91%, and 61.45% for BPNN, GRNN and regression models respectively. | |
keywords: | Cost estimating, artificial neural networks, structural steel buildings | |
full text: | (PDF file, 0.33 MB) | |
citation: | El-Sawah H, Moselhi O (2014). Comparative study in the use of neural networks for order of magnitude cost estimating in construction, ITcon Vol. 19, pg. 462-473, https://www.itcon.org/2014/27 |