Detecting anomalous process behaviour using second generation Artificial Immune Systems

Artificial Immune Systems have been successfully applied to a number of problem domains including fault tolerance and data mining, but have been shown to scale poorly when applied to computer intrusion detection despite the fact that the biological immune system is a very effective anomaly detector. This may be because AIS algorithms have previously been based on the adaptive immune system and biologically-naive models. This paper focuses on describing and testing a more complex and biologically-authentic AIS model, inspired by the interactions between the innate and adaptive immune systems. Its performance on a realistic process anomaly detection problem is shown to be better than standard AIS methods (negative-selection), policy-based anomaly detection methods (systrace), and an alternative innate AIS approach (the DCA). In addition, it is shown that runtime information can be used in combination with system call information to enhance detection capability.