Efficacy Improvement of Anomaly Detection by Using Intelligence Sharing Scheme

Abstract

Computer networks are facing threats of ever-increasing frequency and sophistication. Encryption is becoming the norm in both legitimate and malicious network traffic. Therefore, intrusion detection systems (IDSs) are now required to work efficiently regardless of the encryption. In this study, we propose two new methods to improve the efficacy of the Cisco Cognitive Threat Analytics (CTA) system. In the first method, the efficacy of CTA is improved by sharing of intelligence information across a large number of enterprise networks. In the second method, a four variant-based global reputation model (GRM) is designed by employing an outlier ensemble normalization algorithm in the presence of missing data. Intelligence sharing provides additional information in the intrusion detection process, which is much needed, particularly for analysis of encrypted traffic with inherently low information content. Robustness of the novel outlier ensemble normalization algorithm is also demonstrated. These improvements are measured using both encrypted and non-encrypted network traffic. Results show that the proposed information sharing methods greatly improve the anomaly detection efficacy of malicious network behavior with bad base-line detection efficacy and slightly improve upon the average case.

Keywords

• intrusion detection systems
• global reputation model
• anomaly detection
• intelligence sharing scheme
• CTA-IDS
• machine learning