Neural Networks research demonstrates significant savings in retail forecasting
Published 30 November 2005
The Lancaster Centre for Forecasting has extended research into Forecasting with artificial neural networks. This non-linear, purely data driven class of statistical methods was originally motivated from the functioning of the human brain. Today, they are frequently used by NASA, CIA and other leading research organisations for state-of-the-art applications, e.g. on the Mars Explorer, Terrorist face identification, explosives detection at airports and other classification and data mining projects. Recently, they have generated significant savings in electrical load forecasting, using week of the year, day of the week all through to minute in the hour information, in combination with weather information, bank holidays, external shock etc. A recent evaluation within a domain of similar complexity, forecasting fresh products at UK retail outlets, showed them outperforming all other established methods of multiple linear regression, ARIMAX and Exponential Smoothing. In addition, they provided an intuitive and easy to use modelling process with the potential for large scale automation. Interested ? For more information please contact Sven Crone.
