An association clustering algorithm for can-order policies in the joint replenishment problem

Many studies have shown that the total cost of employing joint replenishment for correlated items is less than the total cost of using single-item replenishment. Savings increase dramatically when the demand between items is closely related. Although the benefits of joint replenishment are significant, it is difficult to define the demand correlation among items, especially when the number of items increases. A large number of items reduces the efficiency and advantage of the multi-item inventory control. To overcome this difficulty, an association clustering algorithm this paper proposes to evaluate the correlated demands among items. The proposed algorithm utilizes the "support" concept in association rule analysis to measure the similarity among items. Based on these measurements a clustering method is developed to group items with close demand in a hierarchal way. The can-order policy is then applied to the optimal clustering result as decided by the proposed performance index. To illustrate the benefits of the proposed association clustering algorithm for replenishment systems, a set of simulations and a sensitivity analysis is conducted. The results of the experiments show that the proposed method outperforms several replenishment models.