171. Machine Learning for Lead-Time Demand Estimation: Addressing Data Sparsity in Spare Parts Inventory Control
Invited abstract in session WC-6: Predictive Analytics: Forecasting I, stream Analytics, Data Science, and Forecasting.
Wednesday, 13:30-15:00Room: H9
Authors (first author is the speaker)
| 1. | Florian E. Sachs
|
| Faculty of Management, Economics and Social Sciences, University of Cologne | |
| 2. | Robin Reiners
|
| Supply Chain Management Area, University of Cologne | |
| 3. | Ulrich Thonemann
|
| Supply Chain Management, Universtiy of Cologne |
Abstract
The cumulative distribution function (CDF) of lead-time demand lies at the core of any inventory control application. Traditionally, the inventory control literature has focused on deriving theoretical distributions for lead-time demand and forecasting future demand. However, recent developments advocate for direct empirical estimation—either by constructing empirical distributions through bootstrapping or by forecasting lead-time demand directly. In this study, we explore the potential of machine learning to enhance CDF estimation, particularly in settings with limited observational data, as is often the case for spare parts. We argue that a density-centric approach, specifically through nonparametric conditional density estimation (CDE), provides a richer and more nuanced representation of the uncertainties surrounding lead-time demand. We benchmark this method against recently proposed bootstrapping and direct forecasting techniques using real-world data, addressing the complexities arising from intermittent lead-time observations. In doing so, we aim to bridge the traditional focus on theoretical lead-time demand distributions with the growing emphasis on direct empirical estimation. Furthermore, by incorporating machine learning methods, we propose a novel and robust approach for estimating the CDF of lead-time demand.
Keywords
- Production and Inventory Systems
- Machine Learning
- Forecasting
Status: accepted
Back to the list of papers