Anomaly Intrusions Detection Based on Support Vector Machines with an Improved Bat Algorithm

The continuous proliferation of more complex and various security threats leads to the conclusion that new solutions are required. Intrusion Detection Systems can be a pertinent solution because they can deal with the large data volumes of logs gathered from the multitude of systems and can even identify new types of attacks if based on anomaly detection. In this paper we propose an IDS model which includes two stages: feature selection with information gain and detection with Support Vector Machines (SVM). A draw-back of SVM is that its performance results are influenced by its user input parameters. Therefore, in order to better the classifier we exploit the advantages of a recent Swarm Intelligence (SI) algorithm, the Bat Algorithm (BA), which we improve by enhancing its randomization with Lévy flights. We test our model for the NSL-KDD dataset and prove that it can outperform the original BA, ABC or the popular PSO.