Forecasting construction demand with a neural network

Forecasting private residential construction demand is an important problem in the real estate industry. While previous research has focused on using multi-linear regression to forecast this demand, we use an artificial neural network to develop the prediction model. The neural network is able to forecast semiannual private residential housing starts in the United States (using only three input variables) with a mean absolute percentage error of 7.6% over a 20-year period. A multi-linear regression model using the same data achieves a much worse mean absolute percentage error of 22% in comparison.