Network intrusion detection by multi-group mathematical programming based classifier

The growing number of computer network attacks or intrusions has caused huge lost to companies, organizations, and governments during the last decade. Intrusion detection, which aims at identifying and predicting network attacks, is a fast developing area that has attracted attention from both industry and academia. Technologies have been developed to detect network intrusions using theories and methods from statistics, machine learning, soft computing, mathematics, and many other fields. We have previously proposed multiple criteria linear programming (MCLP) and multiple criteria nonlinear programming (MCNP) models for two-group intrusion detection. Although these models achieve good results in two-group classification problems, they perform poorly on multi-group situations. In order to solve the problem, we introduce the kernel concept into multiple criteria models in this paper. Experimental results show that the new model provides both high classification accuracies and low false alarm rates in three-group and four-group intrusion detection.