Neural networks for store performance forecasting: an empirical comparison with regression techniques.

Abstract

The paper follows the applied practitioner tradition of location analysis by focusing on the development of tools for store performance assessment. Neural networks have been identified as a highly flexible modeling tool, which have a high, but as yet largely unsubstantiated, potential for managerial problem solving. This paper objectively assesses the suitability for neural networks for store performance modeling by directly comparing them with the tried and tested technique of multiple regression. Based upon data provided by a UK fashion chain store, the research suggests that neural networks offer a highly acceptable forecasting technique which in this case generates marginally better results than multiple regression. It can, therefore, be concluded that neural networks warrant further consideration as an alternative locational analysis tool.

Keywords:

• Neuralnetworks
• Multiple regression
• Location
• Retailing
• Performance