"Techniques For Addressing Extreme Class Imbalance for Artificial Neura" by Colin W. Foley

Date of Award

3-2024

Document Type

Thesis

Degree Name

Master of Science in Operations Research

Department

Department of Operational Sciences

First Advisor

Lance E. Champagne, PhD

Abstract

This research examined the class imbalance problem while training convolutional neural networks (CNN) by applying different techniques to combat this common issue. This research used a modified CIFAR-10 dataset along with a curated aerial image dataset. Methods covered included undersampling, oversampling, synthetic minority oversampling technique, Edited Nearest Neighbors and combinations of the aforementioned methods. This research found that undersampling methods tended to outperform oversampling methods. While undersampling methods showed a decrease in overall accuracy, the increase in minority class prediction performance was promising enough to warrant further investigation.

AFIT Designator

AFIT-ENS-MS-24-M-079

Comments

A 12-month embargo was observed for posting this work on AFIT Scholar.

Distribution Statement A, Approved for Public Release. PA case number on file.

Share

COinS