Predicting Sales with Deep Learning in a Retail Setting

Abstract: Product pricing is an always present issue and there are a number of different traditional pricing strategies that can be applied depending on the situation. With an increasing amount of available data, as well as new improved methods to take advantage of this information, companies are presented with the opportunity to become more data driven in their decision making. The aim of this this thesis is to examine the possibilities of using statistical machine learning methods, more specifically neural networks, to predict what effect price changes have on sales numbers, and to identify what features are of importance when making these predictions. This would allow us to use a more data driven pricing strategy. The work is done in collaboration with Kjell \& Company, a Swedish consumer electronics retailer.  The results of this thesis shows that no predictions regarding sales can be done with any meaningful accuracy using the limited features available at the time of this thesis. More work has to be done in order to identify and quantify more value contributing features. Due to the limitations of the results presented here, no conclusions can be made regarding applying neural networks for these types of problems in general, based on the results of this report. However, the author still believes that it is a promising area of research, and that with a greater domain knowledge, interesting results could be achieved using similar methods.