Application of association rules learning for studying the store history of a large retail chain

Abstract

In this article the problem of goods clustering according to the probability of their joint presence in the transactions of the sale point was considered. The problem was solved by applying the algorithms for the formation of association rules based on real retail store transactions. The applicability of the algorithms of this class was assessed, the Apriori algorithm proposed by R. Agrawal was selected and implemented. Using this algorithm, the possibility of cutting off “random” goods in a cluster to create a consumer basket of the corresponding outlet was studied. It is shown that the most resistant to “random” associations are clusters with a small number of products in the cluster, and the higher the cluster size, the higher the likelihood of noisy search results with “random” goods with a high individual probability of their acquisition (‘support’ indicator). Further development of the proposed method may consist in its distribution to all transactions of the entire trading network and the transition from the tasks of organizing goods within the store (merchandising) to the tasks of forming assortment matrices for stores.