2018 Volume 61 Pages 23-44
We propose a method to analyze the large-scale ID-POS data of supermarkets collected transversely from various management agency. The method reveals store's sale types and customer's purchasing types, and clarifies a purchasing behavior and a tendency of its transitions in each store. At first, we pay attention to distributions of monthly amount of store's sale or customer's purchasing on the product categories, and extract store's sales types and customer's purchasing types, applying self-organizing map (SOM) to these distributions. At this time, we used Box-Cox transformation for an amount of money of a category, because these values are different for wrong number of digits. Then, we clarified the characteristic of the store group, using the composition ratio of the types of customer's purchasing, monthly transitional frequency between types, the number of the participation of the new customer, and so on. As a result, it became clear that a purchasing behavior and the tendency of transition between types of customer's purchasing, such as the rice, meat and fresh fish, were different between the store groups where varied in customer's royalty and age group of customers. It is thought that the proposed method is effective to discover a difference of the constituency of stores.