Sama, Lakshit (2022) Network Intrusion Detection using Deep Learning. Research Master thesis, Victoria University.

Abstract

As network size expands rapidly, network intrusions become more frequent, dynamic, and sophisticated. The topic of how to detect intrusions in such a vast network is crucial and challenging. Due to its intelligent potential, machine learning-based network intrusion detection has recently gained increased attention. Compared to rule-based solutions, machine learning-based solutions, especially those using deep learning, are better capable of identifying network attack variations. In contrast to other application fields, such as image recognition and natural language processing, however, deep learning for network intrusion detection is still in its infancy. It remains to be determined whether it is successful for actual application and if yes, several difficulties need to be studied. This thesis focuses primarily on two challenges associated with deep learning for network intrusion detection. 1) Excessive human intervention in existing machine learning models, high false-positive rate and low accuracy in existing deep learning solutions; 2) Lack of adequate training data in the network intrusion detection sector; We propose a deep learning system (LightGBM, XGBoost, LSTM, and decision tree) to cope with inadequate data and verify it using three datasets. The standard datasets include the NSL-KDD dataset, the UNSW-NB15 dataset, and the CIC-IDS2017 dataset. Each model is selected based on qualities that are likely to increase the framework’s detection. The findings of the suggested framework indicate that all four consistently outperform the state-ofthe- art machine learning-based solutions, demonstrating the efficacy of our thesis-developed design techniques.

Search Google Scholar

Repository staff login