LSTM-based inventory management model using self-attention

Abstract

One of the big concerns of retail companies is to have an effective inventory management system, which considers the sales pattern of products. Consequently, reducing the cost of inventory compared to other retailers has a competitive advantage. However, because of the irregular and stochastic customer demand and different product characteristics, it is considerably difficult to manage a right level of inventory. In this study, we propose a deep learning architecture that finds the optimal balance between shortages and stocks that arise from stochastic customer demands for each product. In a retailer's inventory management problem, there is always a trade-off relationship between a product's shortage and an unsold inventory. In this paper, a loss function is proposed to control the trade-off relationship between such as shortages and the inventory holding. We solve the problem of stochastic demand by applying an LSTM (Long Short-term Memory)-based attention model using the hourly sales volume of products to suggest inventory replenishment orders. Therefore, the core contribution of the paper is to find an optimal inventory policy for products with stochastic demands using the LSTM-based attention model. Moreover, this paper contributes to improving the operational efficiency of retail companies by providing a system that can place inventory replenishment orders while meeting a given target shortage rate.