Forecasting Potential Sales of Bread Products at Stores by Network Embedding

In this work, we study the forecast of the potential sales of out-of-stock products in retail stores using factory shipment data. A precise prediction of the potential sales of out-of-stock products in retail stores is beneficial for both baking factories and retail stores because it optimizes the supply chain by introducing a new product in proper quantity at retail stores, and it also creates new opportunities for baking factories to sell their products to retail stores. This study uses high-dimensional and sparse baking factory shipment data, which are unsuitable for prediction using conventional methods because the data have a high computation time and missing values. We employ a network embedding method, LINE, to derive similar stores based on their sales and predict their potential sales. We confirmed that our proposed method outperforms a simple prediction method (Baseline) and t-SNE for accurate product sales prediction via simulation experiments. We also verified our proposed method's applicability when the forecasting target is expanded to products sold in fewer stores and with lower volume.