Date of Award

9-13-2012

Document Type

Thesis

Degree Name

Master of Science

Department

Department of Electrical and Computer Engineering

First Advisor

Barry E. Mullins, PhD.

Abstract

This research explores the benefits of using commonly-available graphics processing units (GPUs) to perform classification of network traffic using supervised machine learning algorithms. Two full factorial experiments are conducted using a NVIDIA GeForce GTX 280 graphics card. The goal of the first experiment is to create a baseline for the relative performance of the CPU and GPU implementations of artificial neural network (ANN) and support vector machine (SVM) detection methods under varying loads. The goal of the second experiment is to determine the optimal ensemble configuration for classifying processed packet payloads using the GPU anomaly detector. The GPU ANN achieves speedups of 29x over the CPU ANN. The GPU SVM detection method shows training speedups of 85x over the CPU. The GPU ensemble classification system provides accuracies of 99% when classifying network payload traffic, while achieving speedups of 2-15x over the CPU configurations.

AFIT Designator

AFIT-GCO-ENG-12-24

DTIC Accession Number

ADA568667

Share

COinS