The increasing complexity of software and IT systems creates the necessity for research on technologies addressing current key security challenges. To meet security problems in IT infrastructures, a security engineering process has to be established. One crucial factor contributing to a higher level of security is the reliable detection of security vulnerabilities during security tests. We observe the behavior of the system under test and introduce machine learning methods based on derived behavior metrics. This improves the accuracy of the security test result of an automated security testing approach. Reliable automated determination of security failures in security test results increases the security quality of the tested software and avoids costly manual validation.