Fuzzy Kernel Robust Clustering for Anomaly based Intrusion Detection

Abstract

Intrusion Detection Systems (IDS) is a security management system that aim to protect systems from intrusion. Anomaly based detection works based on existence of anomaly behavior that do not conform to a well-defined notation of known normal behavior while misuse detection monitors resources to find intrusion signatures. Unsupervised learning method such as fuzzy clustering method has been widely used to solve IDS problems, especially anomaly based one. The goal is to find patterns on the data to prevent intrusion by detecting anomaly behavior compares to normal. Fuzzy logic based algorithms can be used on several problems that failed to fulfill theoretical assumption such that failed to be modeled analytically. It can also be used at classification problem with incomplete information or problems with unclear boundary classes. In this paper, we use Fuzzy Kernel Robust Clustering algorithm on KDD Cup'99 dataset where we succeeded to classify the items into five clusters, one for normal behavior and the rest are for four types of attacks. The accuracy for every training data is over than 95%, with the lowest one is 95.89% that occur at 10% training dataset, and the highest accuracy is around 98.52% for 80%555 and 90% of training dataset.