Sequential Frequency Vector Based System Call Anomaly Detection

Abstract

Although either of temporal ordering and frequency distribution information embedded in process traces can profile normal process behaviors, but none of ever published schemes uses both of them to detect system call anomaly. This paper claims combining those two kinds of useful information can improve detection performance and firstly proposes sequential frequency vector (SFV) to exploit both temporal ordering and frequency information for system call anomaly detection. Extensive experiments on DARPA-1998 and UNM dataset have substantiated the claim. It is shown that SFV contains richer information and significantly outperforms other techniques in achieving lower false positive rates at 100% detection rate.