A siamese neural network application for sales forecasting of new fashion products using heterogeneous data

In the fashion market, the lack of historical sales data for new products imposes the use of methods based on Stock Keeping Unit (SKU) attributes. Recent works suggest the use of functional data analysis to assign the most accurate sales profiles to each item. An application of siamese neural networks is proposed to perform long-term sales forecasting for new products. A comparative study using benchmark models is conducted on data from a European fashion retailer. This shows that the proposed application can produce valuable item level sales forecasts.