Research of Intrusion Detection Based on an Improved K-means Algorithm

Traditional machine learning methods for intrusiondetection can only detect known attacks since these methodsclassify data based on what they have learned. New attacks areunknown and are difficult to detect because they have notlearned. In this paper, we present an improved k-meansclustering-based intrusion detection method, which trains onunlabeled data in order to detect new attacks. The result ofexperiments run on the KDD Cup 1999 data set shows theimprovement in detection rate and decrease in false positiverate and the ability to detect unknown intrusions.