Machine Learning-Based Security Pattern Recognition Techniques for Code Developers

Software developers represent the bastion of application security against overwhelming cyber-attacks which target all organizations and affect their resilience. As weaknesses may be introduced during process code writing are complex matching different variate skills, most applications launched intrinsically vulnerable. We have advanced our research for a scanner able to use automated learning techniques based on machine algorithms recognize patterns in source code. To make independent programming language, is converted vectorial representation using natural language processing methods, retain semantical traits original at same time reduce dependency lexical structure program. The flaws detection performance ranges accepted by software professionals (recall > 0.94) even when vulnerable samples very low represented dataset (e.g., less than 4% specific CWE dataset). No significant change or adaptation needed under scrutiny. apply this approach detecting Common Weaknesses Enumeration (CWE) vulnerabilities datasets provided NIST (Test suites–NIST Assurance Reference Dataset).