Deep Learning approaches to detect advanced Cyber attacks

Abstract

The deep learning (DL) approach to machine learning (ML) in the field of artificial intelligence (AI) emphasizes high capacity, scalable models that have the capability to learn distributed representations from the input data set. Generality and efficacy of these methodologies in various contextual studies in Cyber Security is shown in this thesis. The neural network models have been adjusted and extended as needed to be more elective all through these studies. The major contributions of this thesis are as follows: Development of a comprehensive database for domain generation algorithm (DGA) generated domain name detection and a novel architecture to improve the overall performance of DGA domain name detection. Development of a deep neural network (DNN) based hybrid intrusion detection alert system which has the capability to analyze the network and host-level activities inside an Ethernet local area network (LAN). Development of a united DL based framework for spam and phishing detection using electronic mail (email), uniform resource locator (URL) and social media data analysis. Development of a DL based approach for secure shell (SSH) traffic analysis, application network traffic classification, malicious traffic classification, and malicious traffic detection. A new proposal of a scalable and hybrid framework, namely ScaleMalNet. This follows two-stage approach, in the first stage the executables file is classified into malware or legitimate using static and dynamic analysis and in second stage the malware executables _le is categorized into corresponding malware family. A similar hybrid DL framework is developed for Android malware analysis and ransomware analysis. This framework is more effective for Android malware and ransomware detection compared to the existing classical ML based methods. Development of a DL based network intrusion detection and DNS based botnet detection framework in the Internet of things (IoT) environment of smart cities.(abstract attached).