We study the usefulness of quantitative data flow analyses for behavior-based malware detection. To this end, we propose a generic model to represent system behavior, i.e. traces of system calls, as sequences of quantifiable data flows that together form a Quantitative Data Flow Graph (QDFG). We operationalize this model in four different ways for highly accurate, robust, and efficient malware detection. We do so by both identifying patterns of known malicious behavior in unknown QDFGs and by profiling malware with graph metrics over QDFGs. Using large data sets, we demonstrate that quantitative data flow analysis yields better detection effectiveness than non-quantitative analysis.
«
We study the usefulness of quantitative data flow analyses for behavior-based malware detection. To this end, we propose a generic model to represent system behavior, i.e. traces of system calls, as sequences of quantifiable data flows that together form a Quantitative Data Flow Graph (QDFG). We operationalize this model in four different ways for highly accurate, robust, and efficient malware detection. We do so by both identifying patterns of known malicious behavior in unknown QDFGs and by pr...
»