Vulnerable Code Detection Using Software Metrics and Machine Learning

摘要: Software metrics are widely-used indicators of software quality and several studies have shown that such can be used to estimate the presence vulnerabilities in code. In this paper, we present a comprehensive experiment study how effective distinguish vulnerable code units from non-vulnerable ones. To end, use machine learning algorithms (Random Forest, Extreme Boosting, Decision Tree, SVM Linear, Radial) extract vulnerability-related knowledge collected source representative projects developed C/C++ (Mozilla Firefox, Linux Kernel, Apache HTTPd, Xen, Glibc). We consider different combinations diverse application scenarios with security concerns (e.g., highly critical or non-critical systems). This contributes understanding whether effectively scenarios, help regard. The main observation is using on top helps indicate relatively high level confidence for security-critical systems (where focus detecting maximum number vulnerabilities, even if false positives reported), but they not helpful low-critical due (that bring an additional development cost frequently affordable).