Time Series Prediction For The Housing Construction Market With The Use Of NARNet

Abstract

The paper discusses the study to develop and tune parameters of a nonlinear autoregressive neural network (NARNet or NARNN) model for predicting the number of housing units. Predictions were made for the housing construction market in Poland, which is a dynamically growing European market. Three stages of the housing construction process have been taken into considerationThe paper discusses the study to develop and tune parameters of a nonlinear autoregressive neural network (NARNet or NARNN) model for predicting the number of housing units. Predictions were made for the housing construction market in Poland, which is a dynamically growing European market. Three stages of the housing construction process have been taken into consideration: permits issued for house construction, houses under construction, and completed new houses. Experimental results have shown that a NARNet model can be a very effective tool in the considered scenario. A network model using the Levenberg-Marquardt backpropagation training function achieved the best model fit, as well as the most accurate one-month predictions.