Early Cost Estimates of Bridge Structures Aided by Artificial Neural Networks

Cost estimates are essential for construction projects success in terms of completion of a project on budget. The estimates that are delivered in the early phase of construction projects are of special importance. The paper presents results of research on applicability of artificial neural networks for early cost estimates of bridge structures. Number of multilayer perceptron networks were investigated as a core of regression models developed to support cost prediction. Basic parameters of bridge structures were used as input values, whereas real life construction costs played the role of expected output values. Data used in the course of the research consisted of information collected for 161 bridge construction projects completed in Poland. One neural network of best performance was selected to be the core of the model with the use of two-step procedure. This network’s structure was 21-2-1 activation functions applied were hyperbolic tangent for hidden layer and linear for output layer. Performance of the model in the light of applied measures such as root mean squared error, mean absolute percentage error and assessment of absolute percentage errors distribution and expectations for early cost estimates is acceptable.