Optimized Demand Forecasting by Cross-Validation

Abstract

Sales forecasting plays an important role in business strategy. An appropriate demand forecasting model is necessary for reducing the cost of storage. At a company level, lowering the warehouse costs and optimizing the value chain is a prominent requirement for an optimum stock management. In this paper a demand forecasting model is built to support the stock management activity of medium enterprises by means of data mining algorithms. SQL server analysis service is used for implementing the demand forecasting model. The paper studies a list of available algorithms that are offered by SQL server analysis service and the performance of the aforementioned algorithms is tested using the cross-validation feature that is provided by SQL server analysis service to optimize the performance of the model. We also aim to explore in our research the ability of RMSE (Root mean Squared Error) to include time series algorithms in the cross-validation phase. The proposed model is tested based on a dataset of a timber export company and the output is used for analysing the performance of the proposed model. The paper reached a group of conclusion and one of most the importance conclusion is neural network algorithms performance was the better in adapting our tested dataset comparing with the other algorithms.