A STUDY OF THE MODIFIED KDD 99 DATASET BY USING CLASSIFIER ENSEMBLES
Mohammed J. Alhaddad · IOSR Journal of Engineering · 2012
Network security has been an important research area.KDD 99 dataset [1 ] has been used to analyze various network security methods.However, it has been shown that this dataset has redundant data points that make the analysis bias for these data points.New modified data sets are proposed that overcome these weaknesses.We carried out the experiments with different classifiers on this datasets to study the applicability of different classification methods for this dataset.Naïve Bayes and decision trees and their ensemble methods are used for this study.We used different performance measures in our study.Results suggest that no single classification method is the best for all types of datasets on all type of performance measures.The comparative performance suggests that classifiers based on decision tree performed better than classifiers based on naïve Bayes.Results also suggest that single decision tree is a good classifier for this data as it has reasonable classification accuracy and less training and testing time.