A fuzzy approach to feature reduction in KDD intrusion detection dataset

Abstract

Analyzing the KDD CUP 99 provides useful information in the development of intrusion detection systems to be used in networks. The classification of records in the KDD dataset into normal and attack records involves mining rules involving the features present in the dataset. Since the KDD dataset contains a huge number of features, mining rules becomes a difficult task. Hence several algorithms have been developed to extract the most relevant set of features that contribute to the accurate classification of records. The selected features should result in the least misclassification rate. This paper presents a fuzzy approach to feature reduction and analyzes the evolved features using classification algorithms in Tanagra. It is found that the algorithm yields a very low misclassification rate when compared to other algorithms.