Sales forecasting by artificial neural networks for the apparel retail chain stores-An application

Sales forecasting with high accuracy is crucial in many industries. Especially, in fast-moving consumer goods, retail and apparel industries, the products are not tailor-made and must be produced and made available in chain stores to the customers, in advance. Therefore, for sales and operations planning, forecast information is required. However, traditionally, time series based forecasting techniques are used that merely consider the seasonality, trend, auto-regressive and cyclic factors. This type of forecasting is not suitable especially in cases where many other factors involved and affect the product sales. In apparel retail industry, special factors such as promotions, special days, weather (temperature), and location of the store may affect the product demands of the chain stores. The unique aspect of this study is that the sales of a product family of the fashion retail chain stores were estimated by means of artificial neural networks, for the first time in the literature. Besides, in this study, new significant factors in forecasting were explored that influence the demand of the chain stores. So, in this study, artificial neural networks are developed and used for sales forecasting of a product family of a real chain store, in Turkey. The stores exist in many cities, and some of the cities have much more stores than the other cities. The city with the highest number of stores was selected and some of the stores in this city chosen among them. The past sales, sales price and promotion data of selected stores are used. In addition, store information, number of customers visiting the store, and weather temperature data are included in the model. Sales are estimated by artificial neural networks. Besides, Regression Analysis was used for forecasting and the results of both techniques were compared. As a result of the study, the most appropriate network structure has been obtained, and a high sales forecasting performance has been reached.