Genetic Algorithm based Refinement Methods for Security Metrics

This paper presents two genetic algorithm based model refinement methods used for vulnerability estimation models. A method presents how model structure refinement is applied to obtain models that estimate the cumulative number of vulnerabilities for a certain product. In this case, empirical observation of similarities between consecutive versions of the product is taken into account. Model structure refinement is presented as procedure. The experimental results show how the method is applied and the results are discussed. The second method uses an aggregated performance indicator as selection criterion in the genetic algorithm. It is shown that simpler models are produced, keeping the quality of estimation comparable with more complex ones. Experimental results confirm the hypotheses