Abstract: New seasonal clothing has these properties: few history sale data and short life cycle, it’s not appropriate to forecast the demand of new seasonal clothing using traditional forecast method based on data sequence serial. However, there exists lots of characteristic information in apparel enterprise’s POS data. A method to abstract characteristic information from POS data was provided to build the sale gene database, which classified the apparel sale data by combining hierarchical clustering and dynamic time wrapping (DTW). And a forecasting method was put forward to calculate the demand of new seasonal clothing based on the constructed gene database, which computed the similarity between a new clothing and a gene of the characteristic database, searched the nearest gene, and compute the demand of the new seasonal clothing based on the information from the nearest gene. The proposed method provides a solution for apparel companies to forecast its demand quantificationally based on its POS data, which can be used to build a quick response system. Experiment was carried out, which abstracted the gene database based on 2011 POS data of a famous apparel enterprise, and the new SKUs sold between 2012-03-01 and 2012-07-31 were used to forecast its demand based on the constructed gene database. And the result proved the effectiveness of the proposed forecasting method.

Key words: demand forecast, gene database, similarity, hierarchical clustering, dynamic time wrapping