Anomaly Based Intrusion Detection Using Meta Ensemble Classifier

Anomaly Based Network Intrusion Detection Systems (ANIDS) mechanisms are largely based on machine learning algorithms and have been found effective in detecting known as well as novel attacks. However, often these algorithms in isolation cannot accurately detect all kinds of attacks and generate lot of false alarms. In this paper, we intend to show that if the power of each of the algorithms are combined and harnessed using an appropriate ensemble method, a significant improvement in detection rate can be achieved. The performance of our meta ensemble classifier was evaluated over several real life intrusion datasets and the benchmark KDD'99 dataset, and the results have been found excellent in comparison to its other competing algorithms.