Intrusion detection in network data is a challenging task because of the velocity of data and data imbalances associated with the domain. This article presents a scalable multilevel hybrid classifier… Click to show full abstract
Intrusion detection in network data is a challenging task because of the velocity of data and data imbalances associated with the domain. This article presents a scalable multilevel hybrid classifier (MLHC) model that can handle huge data and the imbalance associated with network transmission data. The initial level of the proposed model is composed of a hybridized firefly prediction model to identify intrusions. The second-level prediction mechanism is modeled with a Bayesian learner to provide probabilistic predictions. The Bayesian learner is trained with balanced data, thus reducing the effects of imbalance, and only part of the predicted data is passed for secondary prediction, leading to reduced impacts of imbalance and faster and more effective solutions. Experiments were performed on benchmark datasets, namely Knowledge Discovery in Databases (KDD)’99, New Subset and Labeled version of KDD (NSL-KDD), and University of New South Wales (UNSW) datasets, and comparisons were conducted with several recent studies. The results indicate improved performances by up to 37 % in F-measure and 19 % in detection rate, thereby exhibiting the effectiveness and robustness of the proposed model.
               
Click one of the above tabs to view related content.