portal on

forecasting with
artificial neural networks


Citation Analysis







- Free Software CD -



Join our Newsletter





webmaster: Sven F. Crone

Centre for Forecasting
Lancaster University
Management School
Lancaster LA1 4YF
United Kingdom

Tel +44.1524.592991
Fax +44.1524.844885

eMail  sven dot crone (at)
neural-forecasting dot com




Sciatation Analysis of NN in Forecasting

A variety of Literature reviews on business applications of ANNs have been published . Following Adya and Collopy’s survey and evaluation  we conduct a citation analysis at the ISI web of science (review conducted 04/2004) . Our keyword search for neural networks in forecasting yielded 2568 publications, demonstrating an unabated interest an continuous increase of publications on theoretical developments and practical applications since 1988 in a wide range of disciplines, including weather, biological processes, mathematical series and other non-business applications. Fig.1 gives an overview by year.

Fig.1: ISI citation analysis on neural networks for forecasting by year

However, significant differences exist in modelling suitable architectures for business forecasting for each domains, depending on the forecasting objective, application and dataset. Firstly, a variety of ANN publications focus on nominal predictions for classification tasks, which induce different modelling heuristics of little use to derive point predictions through the metric predictions necessary for sales forecasting. In addition, sales forecasting often implies noisy time series with only a few years of monthly or weekly observations, due to structural breaks from launches, relaunches, new product introductions, listing and delistings etc. as opposed to long time series with a high signal-to-noise ratio often found in engineering, physics or artificial time series. As specific heuristics have been developed for distinct applications, we limit our analysis to publications to the sub-domain of sales forecasting, in order to derive robust recommendations for valid and reliable experiments. As a keyword search does not allow a valid separation of publications, we manually evaluated and eliminated all publications related to non-business and non-sales applications and nominal predictions or classifications, e.g. of marketing response rates, market shares, index prices, electrical load forecasting etc. and different application domains in order to derive homogeneous recommendations. We identified additional studies through a bottom-up follow up of citations and additional searches by author name and research group identified in IEEE Xlpore, ScienceDirect, Citeseer, Proquest and ABI/Inform. Our search yielded a total of 171 publications on time series or causal point predictions in business forecasting, in the categories of general publications on point predictions including examples from business, sales, marketing, production and finance.

Fig.2: Distribution of Publications in the Business Forecasting Domain

In the domain of sales forecasting, only 32 publications focussed on sales forecasting of discrete products or services instead of electrical load forecasting etc. This limited number of publications dedicated to the use of ANN in business forecasting for demand planning verifies previous studies and explains the large attention the sparse publications in the field receive. For example, the International Journal of Forecasting ranked publications on NN as the second and fourth most requested papers in 2001 and 2002 .

Home | Neural Associations | Neural Applications | Neural Data Sources | Neural Community | Neural Contacts | Neural Publications | Neural News&Events | Neural Software | Neural Tutorials | Neural Links | Forecasting Principles

©  2002-2005 BI3S-lab - Hamburg, Germany - All rights reserved - Questions, Comments and Enquiries via eMail -  [Impressum & Disclaimer]
The Knowledge Portal on Forecasting with Neural Networks @ www.neural-forecasting.com -