A new simulation framework for intermittent demand forecasting applying classification models

Demand Forecasting is a key to effective inventory management. In forecasting fields, intermittent demand forecasting remains to be a very important but challenging problem. Intermittent demand is characterized by many empty demands, stochastic periods between them, and high variance of non-zero values. These characteristics make intermittent demand forecasting a difficult task, for both parametric and non-parametric approaches. The parametric methods have shown many limitations to provide accurate information. Though non-parametric methods provide better information for decision making than parametric case, they cannot forecast any exact information of point values. This paper proposes a new simulation framework that takes into consideration the correlation structure between demand of assembly and demand of parts, leading to more precise information of point values. In particular, we demonstrate how sub-parts for classification can affect to prediction performance of the overall model via an experiment using artificial data.