Evaluating Different Metric Configurations of an Evolutionary Wrapper for Attack Detection

Detecting various types of attacks is a major problem in cybersecurity. In this paper, we show different configurations of an evolutionary wrapper algorithm for selecting features to classify attacks using a decision tree. We use two metrics for the evaluation function and evolutionary operator acceptance criteria. As part of our experiments, we interchange them and test the effect on the classification quality. Results show that the algorithm is able to guide the classification to accomplish different goals.