Hierarchical forecasting for predicting spare parts demand in the South Korean Navy

Abstract

In the South Korean Navy the demand for many spare parts is infrequent and the volume of items required is irregular. This pattern, known as non-normal demand, makes forecasting difficult. This research uses data obtained from the South Korean Navy to compare the performance of forecasting methods that use hierarchical and direct forecasting strategies for predicting the demand for spare parts. Among various forecasting methods tested, a simple combination of exponential smoothing models, which uses a hierarchical forecasting strategy, was found to minimise forecasting errors and inventory costs. This simple combination forecasting method was generated by a simple combination between an exponential smoothing model with quarterly aggregated data adjusted for linear trend at group level and an exponential smoothing model with monthly aggregated unadjusted data at item level. Logistic regression classification model for predicting the relative performance of alternative forecasting methods (Le. a direct forecasting method vs. a hierarchical forecasting method) by multivariate demand features of spare parts was developed. Logistic regression classification model is generalisable, because it is based on relationships between the relative performance of alternative forecasting methods and demand features. This classification model reduced inventory costs, compared to the results of only using the simple combination forecasting method. This classification model is likely to be a promising model to guide the selection of a forecasting method between alternative forecasting methods for predicting spare parts demand in the South Korean Navy, so that it could maximise the operational availability of weapon systems.