Deep neural networks for sequence based anomaly detection in cyber security

Abstract

Cyber security has become one of the most challenging aspects of modern world digital technology and it has become imperative to minimize and possibly avoid the impact of cybercrimes. Host based intrusion detection systems help to protect systems from various kinds of malicious cyber attacks. One approach is to determine normal behaviour of a system based on sequences of system calls made by processes in the system. The proposed model describes a computationally efficient anomaly based intrusion detection system based on Recurrent Neural Networks. Using Gated Recurrent Units rather than the normal LSTM networks it is possible to obtain a set of comparable results with reduced training times. The incorporation of stacked CNNs with GRUs leads to improved anomaly IDS. Intrusion Detection is based on determining the probability of a particular call sequence occurring from a language model trained on normal call sequences from the ADFA Data set of system call traces. Sequences with a low probability of occurring are classified as an anomaly