Feature Selection in the Corrected KDD-dataset

Abstract

Automation in anomaly detection, which deals with detecting of unknown attacks in the network traffic, has been the focus of research by using data mining techniques in recent years. This study attempts to explore significant features (curse of high dimensionality) in intrusion detection in order to be applied in data mining techniques. Therefore, the existing irrelevant and redundant features are deleted from the dataset resulting faster training and testing process, less resource consumption as well as maintaining high detection rates. The findings were tested on the NSL-KDD datasets (anomaly intrusion datasets) in order to confirm the outcomes.