TWO-STAGE ENSEMBLE LEARNING FOR NIDS MULTICLASS CLASSIFICATION

Abstract

Cyber-attacks have emerged as a pervasive and escalating threat in our increasingly digital world. These attacks target computer systems, networks, and digital infrastructures with malicious intent. Cyber-criminals employ various techniques to exploit vulnerabilities and gain unauthorized access to sensitive information, disrupt operations, or inflict damage. Therefore, cyber-attack mitigation becomes a critical component in safeguarding these assets against such threats. Employing effective mitigation strategies is essential to minimize the impact of cyber-attacks, protect sensitive information, and ensure the integrity of systems and networks. This thesis contributes to the field of cyber-attack classification by investigating various approaches to enhance the performance of multi-class classification in IDSs. The research objectives encompass exploring the performance of different ensemble models, examining the impact of preprocessing techniques using SMOTE for handling imbalanced data, and comparing the effectiveness of two-stage and one-stage classifiers utilizing rotation forest for feature extraction and different ensemble model for second stage. Through rigorous experimentation and analysis, the research identifies the most effective ensemble model, demonstrates the efficacy of SMOTE in addressing imbalanced data, and establishes the superiority of the two-stage classifier approach. Results demonstrate that SMOTE effectively improves the system's ability to detect and classify intrusions accurately, enhancing overall performance and addressing class imbalance. Moreover, the results demonstrated that the adoption of a two-stage classifier consistently improved the overall accuracy, macro average, and weighted average scores compared to the one-stage configuration for various algorithms.

Date of Award	2023
Original language	American English
Awarding Institution	HBKU College of Science and Engineering