Clustering and Sample Selection to Enhance the Performance of the Lamstar Intrusion Detection System

In the present work, it is proposed to enhance the learning capabilities and reduce the training time of a competitive learning LAMSTAR neural network using Clustering and Sample Selection algorithm. KDDCUP99 reduced feature data set (Features reduced by PCA algorithm) is used for training and testing the various classifiers. KDDCUP99 dataset has five classes, DOS, PROBE, NORMAL, U2R, and R2L. The DOS class and Normal class have huge records, which in turn increase the training time of the classifiers. Sample records with high information gain are selected from KDDCUP99 DOS and Normal class using Clustering and sample selection algorithm. The resulting dataset is used to train the LAMSTAR intrusion detection system and its performance compared to other Intrusion Detection classifiers. The results obtained show that the proposed technique performs well in terms of detection rate and training time compared to the results obtained by the classifiers using the full dataset.