NIDS using Machine Learning Classifiers on UNSW-NB15 and KDDCUP99 Datasets

Abstract: The benchmark KDD dataset for intrusion detection system generated a decade ago has become outdated as it does not reflect modern normal behaviors and contemporary synthesized attack activities. In this paper we have used a new UNSW-NB15 data set for NIDS. Pre-processing on this datasets is done using Central Points of attribute values with apriori algorithm to select high ranked feature and remove irrelevant features which causes high false alarm rate. The evaluation of the dataset is performed using machine learning classifiers algorithm: Na�ve Bayes and Logistic Regression. The results show that the decrease in false alarm rate and detection accuracy is improved even after reducing the dataset by eliminating the features and further more reduce in the processing time.

Keywords: Central Point (CP) of attribute values, Apriori, Na�ve Bayes (NB), and Logistic Regression (LR).