Demographic transformation and clustering of transactional data for sales prediction of convenience stores

Reliable retail sales prediction of convenience stores (CVSs) can not only help in making correct purchase decision but also in determining which new products to be launched. Therefore, the main aim of this paper is to propose an enhanced method based on clustering and different abstraction forms of data to forecast the retail sales of CVSs. We use customer type proportion to calculate the similarity of stores to improve the accuracy of clustering. We propose the Extraction of Customer Demographic Characteristics (ECDC) from transactional data as a new approach to solve the problem of the lacking of user information. Driven by domain knowledge, sparse transactional data are aggregated and transformed into ECDC using a rule tree built from shopping habits. The effectiveness of ECDC in clustering of transactional data is demonstrated through its resulting higher accuracy of sales prediction. By experimenting our method with several benchmark methods, our proposed method is found to have an optimal accuracy in forecasting the retail sales of CVSs.