Framework for Deploying IDS Predictive Models

Abstract

AbstractThe focus of this paper is the specification of a particular nascent framework for deploying a system of nonparametric, feed‐forward polynomial neural network (PNN) models that can be used to address intrusion detection system (IDS) deficiencies. The framework is characterized by the formal specification in Z notation of black box operations upon which to base the implementation of an event‐driven Intrusion Detection Support Software System. The deployment of PNN models to realize the operations within IDS3 shows complementary potential with other techniques in improving the ability and cost‐effectiveness of an IDS. The key Intrusion Detection Support Software System operations were derived from a system‐level diagnostic concept called abductive diagnostics that is applicable to IDSs (as well as equipment and medical diagnostics). Several experiments were conducted, which are not detailed in this paper, to train and evaluate PNNs using a commercial machine learning algorithm.