Maintaining cyber security is a complex task, utilizing many levels of network information along with an array of technology. Current practices for combating cyber attacks typically use Intrusion Detection Systems (IDSs) to passively detect and block multi-stage attacks. Because of the speed and force at which a new type of cyber attack can occur, automated detection and response is becoming an apparent necessity. Anomaly-based detection systems, such as statistical-based or clustering algorithms, attempt to address this by analyzing the relative differences in network and host activity. Signature-based IDS systems are typically more accurate for known attacks, but require time and resources for an analyst to update the signature database. This work hypothesizes that the latency from zero-day attack to signature creation can be shortened via anomaly-based algorithms. In particular, the summarizing ability of clustering is leveraged and examined in its applicability of signature creation. This work first investigates a modified density-based clustering algorithm as an IDS, with its strengths and weaknesses identified. Being able to separate malicious from normal activity, the modified algorithm is then applied in a supervised way to signature creation. Lessons learned from the supervised signature creation are then leveraged for the development of unsupervised real-time signature classification. Automating signature creation and classification via clustering turns out satisfactory but with limitations. Density supports for new signatures via clustering can be diluted and lead to misclassification.
Library of Congress Subject Headings
Computer crimes--Mathematical models; Computer crimes--Prevention; Computer security; Computers--Access control; Cluster analysis
Department, Program, or Center
Computer Engineering (KGCOE)
Hendry, Gilbert, "Applicability of clustering to cyber intrusion detection" (2007). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus