Preliminery project cost estimation model using artificial neural networks for public sector office buildings in Sri Lanka

Abstract

Cost estimating is a critical due to incomplete project details and drawings and has become a similar issue in Sri Lanka. Since, cost of a building is impacted by decisions made at the design phase, efficient cost estimation is essential. Therefore novel cost models have identified as simple, understandable and reliable. Thereby, Artificial Neural Networks (ANN) have established having the ability to learn patterns within given inputs and outputs and the end result was developed as the preliminary project cost estimation model for public sector office buildings in Sri Lanka. To accomplish the above aim, the survey approach was selected and semi structured interviews and documentary review were conducted in collecting data. Then training and testing of the Neural Networks (NN) under ten design parameters was carried out using the cost data of twenty office buildings in public sector. The data was applied to the back propagation NNtechnique to attain the optimal NN Architectures. The empirical findings depicts that the success of an ANN is very sensitive to parameters selected in the training process and decreasing learning rate makes Mean Square Error smaller but with considerably larger number of iterations up to certain point. It has been gained good generalization capabilities in testing session achieving accuracy of 90.9% in validation session. Ultimately, NN has provided the best solution to develop a cost estimation model for public sector as accurate, heuristic, flexible and efficient technique.