Toward the Best Feature Model for Network Intrusion Detection Using Stepwise Regression and Support Vector Machine

Abstract

A lightweight network intrusion detection system is more efficient and effective for the real world requirement. Higher performance may result if the insignificant and/or useless features can be eliminated. Stepwise Regression can identify the best feature model from the examined features. In this paper Stepwise Regression and Support Vector Machine are combined to detect network intrusion. Empirical result indicates that using the best feature model obtained from the Stepwise Regression can get nearly the same performance as the full feature set. A comparative study of using different feature selection methods is also shown to prove the usefulness of our approach.