Affiliations: [a] School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing, China | [b] CEDRIC lab, Conservatoire National des Arts et Métiers, Paris, France
Correspondence:
[*]
Corresponding author: Lynda Boukela, School of Computer Science and Engineering, Nanjing University of Science and Technology, 200 Xiaolingwei Street, Nanjing 210094, China. E-mail: lyndaboukela@njust.edu.cn.
Abstract: Nowadays, computers, as well as smart devices, are connected through communication networks making them more vulnerable to attacks. Honeypots are proposed as deception tools but usually used as part of a proactive defense strategy. Hence, this article demonstrates how honeypots data can be analyzed in an active defense strategy. Furthermore, anomaly detection based on unsupervised machine learning techniques allows to build autonomous systems and to detect unknown anomalies without the need for prior knowledge. However, the unsupervised techniques applied for honeypots data analysis do not value the advantages of these tools’ data, particularly the high probability that they include a large number of previously unseen anomalies with unexpected and diverse patterns. Therefore, in the present work, the aim is to improve the unsupervised anomaly detection in honeypots data by varying the data feature subset and the parameterization of the anomaly detection algorithm. To this purpose, an outlier ensemble with LOF (Local Outlier Factor) as a base algorithm is proposed. The ensemble outperforms existing solutions as depicted in the experiments where a detection rate higher than 92% is achieved.
