A Neural-Network-Based Forecasting Method for Ordering Perishable Food in Convenience Stores

In managing convenience stores, placing a balanced order is a critical daily job especially in perishable goods. Making the right decisions in ordering appropriate lot-size can maintain customers' satisfaction; increase store profits reduce the scrap of the perishable food. Neural networks have been proved as an effective pattern recognition and forecasting time series events method. However, existing neural network models need improvements before they can be successfully applied to forecast cold perishable food demand in convenience stores. Sudden changes such as weather may affect the sales volume. This research proposes a neural network model that integrates a dynamic factor to forecast perishable product sales. The experimental results show that this approach is more accurate than conventional time series forecasting models such as the moving average model and autoregressive integrated moving average (ARIMA) model in forecasting perishable food.