The objective of software testing is to stress a program to reveal programming defects. Goal of security testing is, more specifically, to reveal defects that could lead to security problems. Security testing, however, has been mostly interested in the automatic generation of test cases that "try" to reveal a vulnerability, rather than assessing if test cases actually "managed" to reveal vulnerabilities.
In this paper, we cope with the latter problem. We investigated on the feasibility of using tree kernel methods to implement a classifier able to evaluate if a test case revealed a vulnerability, i.e. a security oracle for injection attacks. We compared seven different variants of tree kernel methods in terms of their effectiveness in detecting attacks.
PDF version of the paper.