رسالة ماجستير لنموذج التنبؤ بهجمات الأمن السيبراني عن طريق التعلم البياني (ذكاء اصطناعي)

Prediction Model of Cyber Security Attacks by Graph Learning

Abstract

An insider attack is a well-known type of cybersecurity threat in which the attack is carried out from trust users inside the network. This form of cyber-attack is non-trivial for prediction. The reason is that the threats are predicted through the analysis of user behavior depending on the audit data. The traditional approaches are concerned with learning the user's behavior using independent features that model such statistical correlations over temporal profiles. However, the relations between features are neglected in which it is thought that they can bring strong insight to learning the user's behavior. Thus, this model investigates using a novel neural network structure based on graph attention learning to predict insider threats. The model attempts to map the feature space into a graph-structured space to learn the latent relationships using recurrent neural network architecture to model the anomaly behavior of users, which indicates such a potential attack.

Keywords

Cybersecurity attacks, Deep Learning, Graph learning, Insider Threat, Machine learning