Back propagation in time‐series forecasting

Abstract

One of the major constraints on the use of back propagation neural networks as a practical forecasting tool is the number of training patterns needed. We propose a methodology that reduces the data requirements. The general idea is to use the Box‐Jenkins model in an exploratory phase to identify the 'lag components' of the series, to determine a compact network structure with one input unit for each lag, and then apply the validation procedure. This process minimizes the size of the network and consequently the data required to train the network. The results obtained in eight studies show the potential of the new methodology as an alternative to the traditional time‐series models.

Neural network versus econometric models in forecasting inflation

“…Most uses of ANN in economics have so far been in ®nancial markets, in part because traditional approaches have had low explanatory power and in part because the ANN approach requires abundant data. ANN models have outperformed the traditional time series models in most cases in forecasting stock prices and exchange rates, or in classifying applications such as bond ratings (Ahmadi, 1993;Bosarge, 1993;Kamijo and Tanigavwa, 1993;Sharda and Patil, 1993;Refenes, 1993;Donaldson and Kamstra, 1994;Lachtermacher and Fuller, 1995).…”

Gray relation analysis and multilayer functional link network sales forecasting model for perishable food in convenience store

“…There are no systematic reports on the decision of input nodes. Lachtermacher and Fuller (1995) observed both undesirable effects of more input nodes for one-step ahead forecasting and good effects for multistep prediction. They also found that correct identification of the number of input nodes was more important than the selection of the number of hidden nodes.…”

Work breakdown structure (WBS) development for underground construction

“…Finally, the optimal size of the hidden layer is determined by a search of the possible structures. This search begins with a certain number of nodes, which are the minimum value of a calculated hidden node number based on several rules of thumb that are available in the literature [22][23][24]; in this manner, the model is trained and analyzed. Each time, the number of hidden nodes is increased by one.…”