An Evolutionary Improvement of the Mahalanobis – Taguchi Strategy and Its Application to Intrusion Detection

The Mahalanobis - Taguchi (MT) strategy is a statistical methodology combining various mathematical concepts and is used for diagnosis and classification in multidimensional systems. MT is a very efficient method and has been applied to a wide range of disciplines so far. However, its feature selection phase, which uses experimental designs (orthogonal arrays), is susceptible to improvement. In this paper, we propose a methodology that incorporates MT and a Genetic Algorithm (MT-GA), with the latter being used both for optimizing the feature selection step of MT and for determining the most suitable training set. As an application domain for testing the proposed methodology, we utilized Intrusion Detection Systems (IDS). IDS play an increasingly important role in network security technology nowadays and more and more research is being directed towards building effective diagnostic models. We test the effectiveness of MT-GA by applying it to a well-known intrusion detection dataset and by comparing its performance to that of the typical MT strategy and of other classifiers. The results indicate the benefits of using MT-GA.