Improving Detection Rate in Intrusion Detection Systems Using FCM Clustering to Select Meaningful Landmarks in Incremental Landmark Isomap Algorithm

Abstract

Dimension reduction is crucial when it is applied on intrusion detection systems. Many data mining algorithms have been used for this purpose. For example, manifold learning algorithms, especially Isometric feature mapping (Isomap) have been investigated. Researchers successfully applied Isomap on intrusion detection system as a nonlinear dimension reduction method. But it had some problems such as operation on batch mode and being disabled to handle new data points, additionally, it had computational cost and could not be properly applied on huge datasets. Losing time and reducing speed of detection is another problem of Isomap in intrusion detection systems. Incremental Landmark Isomap which selects landmarks among whole data points has been invented for solving these problems. In this paper, we use FCM as a data reduction method to select meaningful landmarks for Incremental L-Isomap instead of choosing them randomly. This method is implemented and applied on some UCI datasets and also NSLKDD dataset. The results demonstrate higher detection rate for the proposed method, comparing to classical Incremental L-Isomap which chooses landmarks randomly.