A data mining approach to optimise shelf space allocation in consideration of customer purchase and moving behaviours

A good shelf space allocation strategy can help customers easily find product items and dramatically increase store profit. Previous studies generally relied on the space elasticity formula to optimise space allocation models, but space elasticity requires estimates of many parameters, resulting in high costs and frequent errors in the mathematical models. In this study, a three-stage data mining method is proposed for solving the shelf space allocation problem with consideration of both customer purchase and moving behaviours. In the first stage, the customer's purchasing behaviour is derived from records of previous transactions, while moving behaviour is collected through radio frequency identification systems. In the second stage, the A priori algorithm is applied to obtain frequent product association rules from purchase transactions. In addition, the UMSPL algorithm is adopted to derive high-utility mobile sequential patterns from customer mobile transaction sequences. In the third stage, all product items are classified as either major, minor or trivial according to a set of criteria. A Location preference evaluation procedure is then developed to calculate location preference if a minor item is placed at a given section of the store. Based on the location preference matrix, minor items are reassigned to optimal shelves. The experimental results show the proposed method can reassign items to suitable shelves and dramatically increase cross-selling opportunities for major and minor items.