Map Reduce based Deep Hierarchical Extreme Learning Machine with Feature Augmentation for Accurate Intrusion Detection in Big data Environment

Abstract

Intrusion detection in big data environment is very challenging task due to the huge volume and variety of data. Various machine learning and deep learning techniques have been utilized to enhance the performance of intrusion detection systems (IDS) among which the deep learning and ensemble based models have been effective and received immense interest. Still, the time for processing larger data is complex in ensemble and deep learning models. Hierarchical and distributed ensemble models like Extreme Learning Machine (ELM) can be a viable solution for the intrusion detection from larger network data. However, the distributed data blocks of big network data causes large training that is incompatible with the common processing systems under limited time. To resolve these issues, the parallel and distributed processing strategy of the MapReduce framework can be effective. This paper aims at achieving this objective by developing a hybrid deep learning and hierarchical model of ELM using MapReduce framework is proposed in this paper. The proposed MapReduce based Deep Hierarchical Extreme Learning Machine (MR-DHELM) is used in the IDS to handle classification and regression of the larger network intrusion data for identifying the network intrusions. In addition to MR-DHELM, feature augmentation technique is employed using the logarithm marginal density ratios transformation to generate newer, dimension reduced and better-quality training data features to improve the classification. UNSW NB15 dataset is considered for the evaluation of the proposed MR-DHELM with feature augmentation based IDS. Experimental results illustrated that the proposed IDS model detects the network intrusions with higher accuracy of 93.97% within less training/processing time and is significantly better than the state-of-the-art models.

Keywords: Cyber security, Big data, Deep learning, Intrusion Detection Systems, MapReduce, Extreme Learning Machine, Feature augmentation, Deep Hierarchical Extreme Learning Machine.