A Study assessing the application of artificial neural network for preliminary estimation in Sri Lankan building projects

Abstract

Preliminary estimations are prepared at the early stages of every construction project to determine the project's financial feasibility. The Artificial Neural Network (ANN) method is a machine learning method that could also be utilised in preliminary estimation for forecasting and predicting the cost with higher accuracy at a very early project stage. A mixed research approach was used for this research. In the first stage, an ANN model with 11 input attributes was developed, with an obtained accuracy of 89.56% in the validation process. In the second stage, the suitability and applicability of the ANN method for preliminary estimation within the Sri Lankan context were investigated through 10 semi-structured interviews. The frequent use of traditional methods for preliminary estimation practice is widespread. Furthermore, the preferred accuracy is more than 80% in the context. The increased accuracy, time efficiency and usability of the ANN model emphasise the suitability of ANN in the construction industry. However, the insufficiency of the databases within the firms, the lack of programming knowledge, and people’s reluctance to change were identified as challenges. Conversely, initiating a centralised database system within the context, outsourcing the resource requirement to develop the ANN model, and reducing the knowledge gap in the industry regarding modern methods were identified as remedies. Adding location, price fluctuation, and risk uncertainty as input attributes are suggested improvements and modifications for the ANN model, which is the first model with almost 90% accuracy.