A neural network model to forecast national construction output

Construction output forecasts have been routinely made and published to assist construction and construction related firms with their planning processes. They are typically provided by groups of experts, as it is not practical for general construction firms to produce such forecasts for their own uses. The study reported in this paper aims to develop an efficient, practical quantitative method to forecast the national construction demand using economic indicators. Back-propagation neural networks were chosen for this study due to their ability to learn from examples of historical data, to map the relations between various variables without prior knowledge and to generate outputs with relatively high accuracy. Regression models were also developed based on the same data sets. The results were then compared with existing published qualitative forecasts. It was found that the neural network models performed well in forecasting construction output in all three sectors: Housing, Non-housing, and Repair and maintenance. Indeed, in all tests, the neural network models forecasted more accurately than the regression models and were close to the published forecasts. The results show that neural network models can be effectively used as an alternative forecasting tool.