Computer Immunology and Neural Network Models for Masquerader Detection

摘要

A computer immunology model based on finite automata technique is presented in this study to detect masqueraders and identify different users from truncated commands without arguments and enriched commands with arguments. The key mechanism in this immunity process is to distinguish between "self and "non-self". Also probabilistic neural network is applied to classify users into "normal" or "anomalous" by converting the user commands sequence into numerical input feature vectors to the network. Experimental evaluation shows promising results for these two different approaches.