The Empirical Analysis on Proposed Ids Models based on Deep Learning Techniques for Privacy Preserving Cyber Security

Abstract

In AI, the deep learning (DL) method of machine learning (ML) places an emphasis on large-scale, scalable models that can learn distributed representations from their input data. The scope and effectiveness of these techniques are demonstrated in this thesis through a number of case studies pertaining to cyber security. By the end of each study, the neural network models had been fine-tuned and expanded to provide better results. The key arguments presented and discussed in this thesis are as follows: 1) Creating an all-inclusive database for domain name detection using domain generation algorithms (DGAs) and a new architecture to improve DGA domain name detection overall performance. 2) Constructing a hybrid intrusion detection warning system that incorporates deep neural networks (DNNs) to examine host-level and network-level behaviours within an Ethernet LAN. thirdly, analysing data from social media platforms, email, and URLs to create a single DL-based framework for detecting spam and phishing. 4) ScaleMalNet, a novel hybrid framework proposal, is part four. This is a two-step process: first, we use static and dynamic analysis to determine if the executable file is malicious or not. Then, we categorise the malicious executable file into the appropriate malware family. Malware and ransomware analysis for Android is accomplished using a hybrid DL framework that is comparable to this one.