Click on the titles below to expand the information about each abstract.
Viewing 1 results ...
Ahiaga-Dagbui D D, Tokede O, Smith S D and Wamuziri S (2013) A neuro-fuzzy hybrid model for predicting final cost of water infrastructure projects. In: Smith, S D and Ahiaga-Dagbui, D D (Eds.), Proceedings 29th Annual ARCOM Conference, 2-4 September 2013, Reading, UK, Association of Researchers in Construction Management, 181–190.
- Type: Conference Proceedings
- Keywords: artificial neural network, cost estimation, cost modelling, cost overrun, fuzzy set theory
- ISBN/ISSN: 978-0-9552390-7-6
- URL: http://www.arcom.ac.uk/-docs/proceedings/ar2013-0181-0190_Ahiaga-Dagbui_Tokede_Smith_Wamuziri.pdf
Nine out of ten infrastructure projects exceed their initial cost estimates. Accuracy of construction cost estimates remains a contentious area of debate within both academia and industry. Explanations for this have ranged from scope changes, risk and uncertainty, optimism bias, technical and managerial difficulties, suspicions of corruption, lying and insufficient required information for accurate estimation. The capacity for tolerance and imprecise knowledge representation of fuzzy set theory is combined with the learning and generalising capabilities of neural networks to develop neuro-fuzzy hybrid cost models in this paper to predict likely final cost of water infrastructure projects. The will help to increase reliability, flexibility and accuracy of initial cost estimates. Neural networks is first used to develop relative numerical weightings of cost predictors extracted from primary data collected on 98 completed projects. These were then standardised into fuzzy sets to establish a consistent framework for combining the effect of each variable on the overall final cost. A three-point fuzzy lower, upper and mean estimate of likely final cost is generated to provide a tolerance range for final cost rather than the traditional single point estimate. The performance of the final models ranged from 3.3% underestimation to 1.6 % overestimation. The best models however averaged an error of 0.6% underestimation and 0.8% overestimation of final cost of the project. The results are now being extended to a larger database of about 4500 projects in collaboration with an industry partner.