Abstract:

Demand forecast is the basis of business operation for a company. Forecast accuracy has a great effect on safety inventory, profit and competitiveness. In this paper, support vector machine (SVM) is used to forecast the demand of a product category including multiple brands. Various factors that affect the product demand such as seasonal and promotional factors are taken into consideration. Data used in model training and assessment are generated from the demand function of this product. Different forecast models such as regression model, double exponent smooth model, Winter model and radial basis function neural network model are used for comparison and evaluation. The results show that accuracy of SVM is superior to other models, which can significantly reduce the inventory level. Therefore SVM is shown to be an effective model for demand forecast.

Key words: demand forecast; support vector machine (SVM); statistical forecast method; neural network