LUMPY DEMAND FORECASTING USING LINEAR EXPONENTIAL SMOOTHING, ARTIFICIAL NEURAL NETWORK, AND BOOTSTRAP

Abstract

Lumpy demand represents the circumstances when a demand for an item has a large proportion of periods having zero demand. This certain situation makes the time series methods might become inappropriate due to the model’s inability to capture the demand pattern. This research aims to compare several forecasting methods for lumpy demand that is represented by the demand of spare part. Three forecasting methods are chosen; Linear Exponential Smoothing (LES), Artificial Neural Network (ANN), and Bootstrap. The Mean Absolute Scaled Error (MASE) is used to measure the forecast performance. In order to gain more understanding on the effect of the forecasting method on spare parts inventory management, inventory simulation using oil and gas company’s data is then conducted. Two inventory parameters; average inventory and service level; are used to measure the performance. The result shows that ANN is found to be the best method for spare part forecasting with MASE of 0,761. From the inventory simulation, the appropriate forecasting method on spare parts inventory management is able to reduce average inventory by 11,9% and increase service level by 10,7%. This result justifies that selecting the appropriate forecasting method is one of the ways to achieve spare part inventory management’s goal.