Determination of order quantity for perishable products by using the support vector machine

Abstract

Daily meal boxes are perishable goods which, if not sold within their limited lifetimes, will be discarded. Convenience stores generally make their order decisions according to point of sale (POS) analysis. However, due to the effects of uncertain factors, such as the weather, temperature, the number of customers, and promotion activity of substitute products, the order quantity based on POS may not actually match with real demand, especially for perishable items. In this study, a novel warning system is established by employing the support vector machine (SVM) to modify the order quantity; and the Taguchi method is applied to determine the optimal portfolio of the factors that may influence the prediction accuracy of the SVM. Using actual data from a convenience store, which is a part of the President Chain Store Corporation in Taiwan, the prediction accuracy of the warning system was evaluated. Through numerical experiments, that the proposed methodology can significantly raise the profit is confirmed.