Hybrid Machine Learning Algorithms for Network Anomaly Detection: A Soft Voting Ensemble Approach

Abstract

Network anomaly detection plays a vital role in maintaining the security and integrity of computer networks in today’s dynamic world, as they face new and different cyber attacks every day. The traditional approach of signature-based anomaly detection does not detect some anomalies, which shows the necessity for a new system that can adapt better using a more effective machine learning approach. In this paper, a new hybrid machine learning method is presented for the purpose of detecting network anomalies based on NSL-KDD data sets. Three well-known machine learning algorithms, which are; Support Vector Machine, Random Forest, and Artificial Neural Network, are combined by using a soft voting technique in order to detect network anomalies more effectively. Different scenarios were tested, from single algorithms to combinations of these algorithms. As per the experimentation results, Scenario 7, which included all three algorithms, yielded the best performance with 75.767% accuracy.