Date of Award

3-23-2018

Document Type

Thesis

Degree Name

Master of Science in Operations Research

Department

Department of Operational Sciences

First Advisor

Bradley C. Boehmke, PhD.

Abstract

Air Force Installation Contracting Agency manages nearly 18 percent of total Air Force spend, equating to approximately 57 billion dollars. To improve strategic sourcing, the organization is beginning to categorize installation-support spend and assign accountable portfolio managers to respective spend categories. A critical task in this new strategic environment includes the appropriate categorization of Air Force contracts into newly created, manageable spend categories. It has been recognized that current composite categories have the opportunity to be further distinguished into sub-categories leveraging text analytics on the contract descriptions. Furthermore, upon establishing newly constructed categories, future contracts must be classified into these newly constructed categories in order to be strategically managed. This research proposes a methodological framework for using Latent Dirichlet Allocation to sculpt categories from the natural distribution of contract topics, and assesses the appropriateness of supervised learning classification algorithms such as Support Vector Machines, Random Forests, and Weighted K-Nearest Neighbors models to classify future unseen contracts. The results suggest a significant improvement in modeled spend categories over the existing categories, facilitating more accurate classification of unseen contracts into their respective sub-categories.

AFIT Designator

AFIT-ENS-MS-18-M-161

DTIC Accession Number

AD1056419

Download

Share

COinS