Development of the time-series forecasting model by an artificial neural network in the CVS ordering system

As to manager the convenience stores (CVS), how to place an accurate order is a quite critical job in daily works, especially on the perishable goods. Making a right decision to order an appropriate lot-size can reduce the scrap of the perishable food and maintain the customers' satisfaction at the same time such that the profit of the CVS can be increased. Recently, neural network has shown to be an effective method in many research areas. However, the existing neural network models still need some improvements before they can successfully be applied. In this study, we present an artificial neural network (ANN) model by using the past sales data in two days ago to seven days ago to take as the input data for forecasting the sales then compared with the famous time series forecasting autoregoressive integrated moving average (ARIMA) model. The results show that the ANN forecasting model is better than the current CVS ordering system in forecasting the 30, 60, 90 and 120 days sales. Besides the ANN forecasting model which use the 5 similar to 7 lag days data to be the input data is better than ARIMA model in long-term forecasting in 90 and 120 days.