Academic Commons


Data Sanitization: Improving the Forensic Utility of Anomaly Detection Systems

Cretu, Gabriela F.; Stavrou, Angelos; Stolfo, Salvatore; Keromytis, Angelos D.

Anomaly Detection (AD) sensors have become an invaluable tool for forensic analysis and intrusion detection. Unfortunately, the detection performance of all learning-based ADs depends heavily on the quality of the training data. In this paper, we extend the training phase of an AD to include a sanitization phase. This phase significantly improves the quality of unlabeled training data by making them as "attack-free" as possible in the absence of absolute ground truth. Our approach is agnostic to the underlying AD, boosting its performance based solely on training-data sanitization. Our approach is to generate multiple AD models for content-based AD sensors trained on small slices of the training data. These AD "micro-models" are used to test the training data, producing alerts for each training input. We employ voting techniques to determine which of these training items are likely attacks. Our preliminary results show that sanitization increases 0-day attack detection while in most cases reducing the false positive rate. We analyze the performance gains when we deploy sanitized versus unsanitized AD systems in combination with expensive hostbased attack-detection systems. Finally, we show that our system incurs only an initial modest cost, which can be amortized over time during online operation.



More About This Work

Academic Units
Computer Science
Department of Computer Science, Columbia University
Columbia University Computer Science Technical Reports, CUCS-011-07
Published Here
April 28, 2011