PERFORMANCE COMPARISON OF INTRUSION DETECTION SYSTEM CLASSIFIERS USING VARIOUS FEATURE REDUCTION TECHNIQUES

In this paper, we compare the performance of Intrus ion Detection System Classifiers using various feature reduction techniques. To enhance the learni ng capabilities and reduce the computational intens ity of competitive learning neural network classifiers, di fferent dimension reduction techniques have been pr oposed. These include: Principal Component Analysis, Linear Discriminant Analysis, Independent Component Analy sis. Many Intrusion Detection Systems are based on neura l networks. However, they are computationally very demanding . In order to mitigate this problem, dime nsion reduction techniques are applied to a given d ataset to extract important features. In the proposed researc h various classifiers are applied to the reduced fe atur dataset and their performance is compared. On the basis of these results, a technique is proposed which perfor ms exceptionally well, in terms of both accuracy and c omputation time. When applied to the KDDCUP99 reduc ed feature dataset, this technique performs better tha n a standard learning schema based on the full feat ured dataset.