Predicting Changes in Construction Cost Indexes Using Neural Networks

TLDR

Construction cost indexes compare period‑to‑period cost changes for a fixed quantity of goods or services. Back‑propagation neural‑network models were trained on macroeconomic data from 1967‑1991, using recent index trends, prime lending rate, housing starts, and month of year as inputs, to forecast one‑ and six‑month changes in the ENR construction cost index and were compared with exponential smoothing and linear regression predictions. The neural network predictions had greater error than exponential smoothing or linear regression, indicating that cost index movements are complex and cannot be accurately predicted by a back‑propagation neural‑network model.

Abstract

Construction cost indexes provide a comparison of cost changes from period to period for a fixed quantity of goods or services. Back‐propagation neural‐network models have been developed to predict the change in the ENR construction cost index for one month and six months ahead. A training set of macroeconomic data was developed for the period from 1967 to 1991. The neural‐network models use inputs including recent trends in the index, the prime lending rate, housing starts, and the month of the year. Output from the neural‐network models was compared with predictions made by exponential smoothing and simple linear regression. The prediction produced by the neural network gave a greater error than either exponential smoothing or linear regression. It can be concluded that the movement of the cost indexes is a complex problem that cannot be predicted accurately by a back‐propagation neural‐network model.