Automated Generation of Attack Graphs Using NVD

Today's computer networks are prone to sophisticated multi-step, multi-host attacks. Common approaches of identifying vulnerabilities and analyzing the security of such networks with naive methods such as counting the number of vulnerabilities, or examining the vulnerabilities independently produces incomprehensive and limited security assessment results. On the other hand, attack graphs generated from the identified vulnerabilities at a network illustrate security risks via attack paths that are not apparent with the results of the primitive approaches. One common technique of generating attack graphs requires well established definitions and data of prerequisites and postconditions relating to the known vulnerabilities. A number of works suggest prerequisite and postcondition categorization schemes for software vulnerabilities. However, generating them in an automated way is an open issue. In this paper, we first define a model that evolves over the previous works to depict the requirements of exploiting vulnerabilities for generating attack graphs. Then we describe and compare the results of two different novel approaches (rule-based and machine learning-employed) that we propose for generating attacker privilege fields as prerequisites and postconditions from the National Vulnerability Database (NVD) in an automated way. We observe that prerequisite and postcondition privileges can be generated with overall accuracy rates of 88,8 % and 95,7 % with rule-based and machine learning-employed (Multilayer Perceptron) models respectively.