Prediction of perishable goods deliveries by GRU neural networks for reduction of logistics costs

Forecasting of demands of perishable goods is cru-cial in planning production schedules to satisfy customer needs on time, and to lower the profit losses of over or under stocks. Collaboration with one of local the supermarket chains provided a reasonable foundation for this academic study to optimize the forecasting of deliveries of perishable goods for food supply chains. By carefully analyzing its logistics operations and real-time data of short-shelf life product deliveries, it is discovered that the current supply management of the stores is solely based on prior managerial experiences, taking into consideration the spoilage, stock-out rates, and holiday seasons. Sudden change in demand causes problems to managers who struggle with keeping up with unpredictable frequency, type, and quantity of goods delivered to a particular place from an assigned warehouse. The paper presents a methodology for reliable planning and scheduling of orders of perishable goods, enabling planners to construct delivery schedules having a low expected total cost. This study aims to implement artificial intelligence where the demand for perishable goods can be predicted a few days in advance, also capable to cope with sudden changes. For that, the Gated Recurrent Unit recurrent neural networks are providing 81.3% average accuracy for observed 10 delivery points. Accurate prediction of demand results in delivering fresher products, which translates into economic benefits in terms of a higher product price.