Applying prediction methods for nonstationary time series from an Distributing Company

摘要

This work addresses the challenge of supplying the stock of an organization based on peculiar behavior of each unit sale. For that, sales data from a single product of an Distributing Company, whose performance is derived by the efforts of seven subsidiary units, are used for performance analysis of some methods of prediction in this context. The Poisson method was used to calculate the probability of meeting all demand for the stock desired. The Moving Averages (MM), Holt-Winters (HW) and Elman Neural Network (ANNe) were applied as prediction methods. Based on the mean square error performance the methods were compared. As a result, the HW method performed best, followed by MM and ANNe.