A Filter Based Feature Selection Approach in MSVM Using DCA and Its Application in Network Intrusion Detection

摘要

We develop a filter based feature selection approach in Multi-classification by optimizing the so called Generic Feature Selection (GeFS) measure and then using Multi Support Vector Machine (MSVM) classifiers. The problem is first formulated as a polynomial mixed 0-1 fractional programming and then equivalently transformed into a mixed 0-1 linear programming (M01LP) problem. DCA (Difference of Convex functions Algorithm), an innovative approach in nonconvex programming framework, is investigated to solve the M01LP problem. The proposed algorithm is applied on Intrusion Detection Systems (IDSs) and experiments are conducted through the benchmark KDD Cup 1999 dataset which contains millions of connection records audited and includes a wide variety of intrusions simulated in a military network environment. We compare our method with an embedded based method for MSVM using l_2 - l_0 regularizer. Preliminary numerical results show that the proposed algorithm is comparable with l_2 - l_0 regularizer MSVM on the ability of classification but requires less computation.