A Novel Approach of Intermittent Demand Prediction in Industrial Domain

Abstract

Intermittent demand, one of the categories of unusual demand, such as demand of spare parts of industry equipment management, seasonal and short life cycle products, is characterized by infrequent demand arrivals and variable demand sizes, which is difficult to predict. To solve this problem, a novel approach was proposed in this paper to predict intermittent demand. It provided mechanism to forecast demand arrival time with demand values at the same time when demand occurs. It firstly applied Decision Tree to predict a 0-1 binary values of demand arrival time. Meanwhile Neural Network was built to predict demand values. Then the time points of these two results were matched. The demand values for those moments when predicted demand that did not occur were replaced by zero. Finally we applied this approach to predict sparse demand of products, the results showed that the predicting accuracy was superior to traditional Back-propagation Neural Network through comparison.