Date of Award

2022

Degree Name

Mathematics

College

College of Liberal Arts

Type of Degree

M.A.

Document Type

Thesis

First Advisor

Dr. Raid Al-Aqtash, Committee Chairperson

Second Advisor

Dr. Anna Mummert

Third Advisor

Dr. Laura Adkins

Abstract

Cyberattack is a never-ending war that has greatly threatened secured information systems. The development of automated and intelligent systems provides more computing power to hackers to steal information, destroy data or system resources, and has raised global security issues. Statistical and Data mining tools have received continuous research and improvements. These tools have been adopted to create sophisticated intrusion detection systems that help information systems mitigate and defend against cyberattacks. However, the advancement in technology and accessibility of information makes more identifiable elements that can be used to gain unauthorized access to systems and resources. Data mining and classification tools such as K-Nearest Neighbors, Support vector machines, and Decision trees, among others, have been improved over time and used to build models for intrusion detection systems. This enables information systems, internet-connected devices, or devices running on a computer network to gain immunity against cyberattacks. However, these classification models hit some limitations as the sample size of data increases. Neural networks is an artificial intelligence tool that has been in active research over recent years. It has proven to handle big data and understand complex relationships better than the previous classification methods. Recent studies have demonstrated to build better models by showing better accuracy for intrusion detection systems using neural networks. In this thesis, we use a class of neural networks known as Self-Normalizing Neural Networks, which implements a scaled exponential linear unit activation function (SELU) developed by Klambauer et al. [12], to build a predictive model to detect cyberattacks against normal network traffic or connections using classification, in the KDD CUP 99 dataset from the Third International Knowledge Discovery and Data Mining Tools Competition, that was held in 1999. The accuracy and precision of the self-normalizing neural networks is compared with that of the k-nearest neighbors and support vector machines. The self-normalizing neural network appears to perform better. It is an excellent classifier for denial-of-service attacks, probe attacks, and user-to-root attacks while efficiently predicting normal connection. The result in this thesis is also compared with existing literature which appears to perform better.

Subject(s)

Cyberterrorism.

Data mining – Statistical methods.

Data mining – Implements.

Support vector machines.

Decision trees.

Machine learning.

Neural networks (computer science) – Research.

Share

COinS