A backpropagation neural network for sales forecasting

This study uses a backpropagation neural network (BPN) to forecast future sales volumes of a food product for a large Victorian food wholesaler. The study compares the results obtained using different parameter values, and discusses network performance. The BNP results are compared with the methods currently used by the company which involve trend and market analysis using a simple linear regression model. The BPN appears to give better forecasts than the statistical methods. Specific factors such as advertising, and competition from other competitors were not included. It is believed that some of these factors may be important. The results obtained suggest that the BPN model may provide a useful tool for generating sales forecasts, however poor selection of parameter settings can lead to slow convergence and/or incorrect output