Analysis of NSL-KDD Dataset for Classification of Attacks Based on Intrusion Detection System Using Binary Logistics and Multinomial Logistics

Abstract

At present, the intrusion detection system is the most developed trend in society. The intrusion detection system acts as a defense tool to detect security attacks which has been increasing steadily in recent years. Therefore, global information security is a very serious problem. As the types of attacks that emerge are constantly changing, there is a need to develop adaptive and flexible security features. Selection feature is one of the focuses of research on data mining for datasets that have relatively many attributes. In this study, the author tries to analyze the NSL-KDD data set with the selected attributes classified in two ways, namely binary classification (attack or not attack) and five classification classes using multinomial logistics, namely Dos, R2L, U2R, Probe and Normal. The results showed that the NSL-KDD dataset for the classification of attacks on the Intrusion Detection System (IDS) using binary logistics can increase the classification accuracy to 92.3% and 91.7% for datasets with multinomial logistics.