An Evaluation Framework for Publications on Artificial Neural networks in Sales Forecasting

Although artificial neural networks (ANN) promise superior performance in forecasting theory, they are not yet an established method in business practice. The vast degrees of freedom to parameterize ANNs have lead to countless heuristic approaches to simplify modeling, training, network selection and evaluation implemented with varying success. Consequently, a systematic evaluation is required in order to identify successful heuristics and derive sound guidelines to ANN modeling from publications. As each forecasting domain imposes different heuristics for classification or point prediction on specific datasets, a literature review is conducted, identifying 2538 publications within the domain of ANN forecasting but only 32 of them applicable to the domain of sales forecasting. The identified publications are evaluated through a framework regarding their validity and reliability in experiment design and documentation, in order to promote superior publications, derive recommendations for future experiments and possibly identify gaps in current research and practice.