Dynamic Weighted Voting Classifier for Network Intrusion Detection

Abstract

Network security is important for countries, companies, and other governments. Network intrusion detection becomes more and more critical for numerous applications. In network intrusion detection, ensembles are often used to improve the performance of single classifiers. However, how to assign weights for the different classifiers is a problem. Instead of using the simple majority voting method, multiple ways to assign global weights are introduced to achieve better performance. In this paper, a new way of dynamically updating weights while predicting is proposed and applied to the classification problem on the UNSW-2015 dataset. The result shows that the dynamic weighted voting classifier performs better than the fixed weighted voting and simple majority rule voting in general.