Predicting Order Status using XGBoost

Author

Kegan J. Penovich, Utah State UniversityFollow

Date of Award

8-2022

Degree Type

Report

Degree Name

Master of Mathematics (MMath)

Department

Mathematics and Statistics

Committee Chair(s)

Joseph Koebbe

Committee

Joseph Koebbe

Committee

Tyler Brough

Committee

Kevin Moon

Abstract

Invista, a Koch subsidiary, is a multinational producer of fibers, resins, and intermediaries, particularly nylon. To keep the company operating required them to take over 1.5 million orders over the course of - years, less than a third of which arrived on-time. Orders arriving other than when expected can cause many problems for any company. While arriving late is a clear problem, it also troublesome for them to arrive early. In the face of this, it becomes important to be able to tell a-priori if an order will arrive on-time or not.

To address this problem, we made use of those 1.5 million orders to try and learn how to predict if an order would be on-time or not. There are many methods for doing so and we tried three approaches: Neural-Networks, Gradient Boosting, and Time series. In the end we found the Gradient Boosting algorithm worked the best. We utilized the popular XGBoost framework of Gradient boosting. This was made further appealing by the company having utilized this algorithm before.

Recommended Citation

Penovich, Kegan J., "Predicting Order Status using XGBoost" (2022). All Graduate Plan B and other Reports, Spring 1920 to Spring 2023. 1655.
https://digitalcommons.usu.edu/gradreports/1655