Accession Number:

AD1141471

Title:

Backorder Prediction

Descriptive Note:

[Technical Report, Technical Report]

Corporate Author:

IBM Corporation

Report Date:

2021-02-26

Pagination or Media Count:

54

Abstract:

This research project was conducted by IBM Corporation from July 2020 - February 2021. The Defense Logistics Agency DLA awarded IBM a contract to research the development of predictive models using machine learning algorithms to predict item backorders and to research the factors contributing to the potential backorder. IBM researched two supervised machine learning approaches 1. Classical learning models widely used and easily interpretable algorithms such as decision trees and linear regression techniques that learn to map some input to an associated output through statistical methods. 2. Artificial Neural Network models computational systems loosely based on biological neural networks that allow for more complex non-linear relationships between the response variable and its predictors. IBM demonstrated that machine learning models using a cleaned, limited, and curated dataset can forecast monthly backordered versus in-stock items, predict the total number of backorders at the item level, and identify important correlated features for each month in the forecast, up to 12 forward looking months. This research provides an entry-point to enhance these initial machine learning models through integrating additional data sources, scaling to larger segments of data, and gaining deeper insight into contributing backorder factors. This research highlights the foresight necessary for DLA to act on anticipated backorders, increasing product availability to its customers.

Subject Categories:

  • Cybernetics

Distribution Statement:

[A, Approved For Public Release]