Ensemble of Feature Chains for Anomaly Detection

Along with recent technological advances more and more new threats and advanced cyber-attacks appear unexpectedly. Developing methods which allow for identification and defense against such unknown threats is of great importance. In this paper we propose new ensemble method (which improves over the known cross-feature analysis, CFA, technique) allowing solving anomaly detection problem in semi-supervised settings using well established supervised learning algorithms. Theoretical correctness of the proposed method is demonstrated. Empirical evaluation results on Android malware datasets demonstrate effectiveness of the proposed approach and its superiority against the original CFA detection method.