A Deep Learning Approach for Network Intrusion Detection Using Non-symmetric Auto-encoder

Abstract

In the present era, network intrusion detection system (NIDS) has become the backbone of network security. But providing complete security to online resources from network intrusions is still lacking. Various approaches have been framed to enable effective detection of network-based intrusions by NIDS. However, various problems like high false alarm rates and poor detection accuracy for minority classes have been encountered during deployment of these systems. To overcome these shortcomings of existing systems, the proposed work provides an approach via a deep learning method used for detecting network intrusions efficiently and effectively. The suggested non-symmetric deep auto-encoder (NDAE-IDS) is detailed for two different scenarios, viz. for binary class and multi-class problems. The proposed approach is executed on GPU-enabled TensorFlow and assessed with the benchmark data set KDD Cup’99. Performance evaluation results obtained indicate that the proposed approach detects network intrusions more accurately and effectively as compared to conventional machine learning methods. Comparative analysis with two recent deep learning approaches using auto-encoders shows that the proposed approach attains better results for detection rate, accuracy and false alarm rate metrics for KDD Cup’99 dataset.KeywordsAsymmetric auto-encoderAnomaly-based IDSDeep learningKDD Cup’99Network security