In this article, we will address the problem of automatic classification of security vulnerabilities by detecting those that target critical infrastructures from those who are not, through the classification of vulnerabilities existing in the CVE database. The existing issue is the complexity to detect vulnerabilities targeting only critical infrastructure, based uniquely on the CVE data base, reason why it is necessary to use data analysis, domain knowledge, NLP, and machine learning techniques such as clustering, labeling, word embedding, classification, modeling and evaluation, in order to represent an efficient solution for this problem.