Risk Modeling on Cyber Vulnerability Graphs

Abstract

Cyber security is a field which gets an increasing amount of attention. Record Future is a company providing services to aid and understand this field. An important part of their service is the Security Intelligence Graph (SIG), describing entities in the cyber threat world and how they relate to one and other. This master’s thesis proposes a k-neighborhood graph approach for estimating security risks regarding Internet Domains which are present on the SIG. In order to classify Internet Domains their k-neighborhood graphs, on SIG, are mapped to a feature space using Weisfeiler-Lehman graph kernels. Two class and three class classifiers are implemented using support vector machines and logistic regression models. The results suggest that this approach works well for the two class classifier, where the models achieve accuracies around 0.96 and F1 scores around 0.89. The results are less promising for the multiclass models.