A multi-target approach to estimate software vulnerability characteristics and severity scores.

Abstract Software vulnerabilities constitute a great risk for the IT community. The specification of the vulnerability characteristics is a crucial procedure, since the characteristics are used as input for a plethora of vulnerability scoring systems. Currently, the determination of the specific characteristics -that represent each vulnerability- is a process that is performed manually by the IT security experts. However, the vulnerability description can be very informative and useful to predict vulnerability characteristics. The primary goal of this research is the enhancement, the acceleration and the support of the manual procedure of the vulnerability characteristic assignment. To achieve this goal, a model, which combines texts analysis and multi-target classification techniques was developed. This model estimates the vulnerability characteristics and subsequently, calculates the vulnerability severity scores from the predicted characteristics. To perform the present research, a dataset that contains 99,091 records from a large -publicly available- vulnerability database was used. The results are encouraging, since they show accuracy in the prediction of the vulnerability characteristics and scores.