Improved K-Means Clustering Algorithm to Detect Network Intrusions

Intrusion Detection Systems are integral part of system's security and are one of the fastest technologies within the security space. Various approaches to Intrusion Detection are currently being used, but they are relatively ineffective. The major problem in intrusion detection system research is the speed of detection - good measure of performance since they measure what percentage of intrusions the system is able to detect. The basic K-Means clustering algorithm is inefficient on high data sets due to its unbounded convergence of cluster centroids. So for removing this problem we have adopted an improved optimum cluster initialization algorithm to obtain effective and efficient crisp clusters in Intrusion Detection System. The technique is tested using multitude of background knowledge sets in DARPA network traffic datasets.