The following research is about using Neural Networks to identify risky areas of software code (sources. Not a compiled binary) that might be exploited in the future. It includes the use of vulnerable files from CVE Details (www.cvedetails.com) as an example of a risky code.

The algorithm developed breaks the files into several logic blocks that served as a training set for the network using known concepts of lexical analysis suck as tokenization and sub tokenization And for the correct desired outcome it uses the CVSS score of the file.

After some training of the network the algorithm produced some interesting results. And was then tested on specifically written risky code.