According to a 2022 report released by the Department of Defense (DoD) Inspector General “The inability to produce accurate and timely forecasts of joint logistics needs created an unmitigated risk to the DoD’s ability to plan and logistically support operations and contingencies.”(DoD-IG, 2022) This assessment is one in a series of reports surrounding the ability for the DoD to perform material forecasting. In 2015 U.S. Government Accountability Office (GAO) report, DoD supply chain management has been classified as a high-risk area since 1990, due in part to weaknesses in accurately forecasting the demand for spare parts. This paper ranks multiple machine learning and statistical approaches to determine the most accurate method for forecasting repairable part demand in U.S. Army simulators based upon Root Mean Squared Error (RMSE) and the Mean Absolute Error (MAE) values. The motivation for this paper is there are relatively few studies that compare the performance of statistical and machine learning forecasting methods in performing intermittent demand forecasting of repair parts for military systems (Ahmed et al., 2010; Makridakis et al., 2018; Spiliotis et al., 2020). Of the few comparative studies that have been performed, several of them place computational limitations on the machine learning methods. This paper removes the computational limitation placed against machine learning methods when comparing the achieved accuracy to statistical models. Additionally, this paper addresses the practical issue of assessing forecasting accuracy while dealing with an inventory suffering from part obsolescence, which is common in military systems (Ruud H. Teunter et al., 2011). The results of this paper found that Support Vector Regression (SVR) is the most accurate method when forecasting repairable and consumable part demand for simulators classified as NMC, PMC, and FMC. However, K-Nearest Neighbor Regression (KNNR) tied the accuracy of SVR when forecasting material demand when the system was in a PMC state.
Keywords
MACHINE LEARNING
Additional Keywords
Time series forecasting, Intermittent Demand Forecasting, Machine Learning, Forecasting Accuracy, Neural Networks, Forecast Validation