Proposed Mitigation Framework for the Internet of Insecure Things

Abstract

Intrusion detection systems IDS are increasingly utilizing machine learning methods. IDSs are important tools for ensuring the security of network data and resources. The Internet of Things (IoT) is an expanding network of intelligent machines and sensors. However, they are vulnerable to attackers because of the ubiquitous and extensive IoT networks. Datasets from intrusion detection systems (IDS) have been analyzed deep learning methods such as Bidirectional long-short term memory (BiLSTM). This research presents an BiLSTM intrusion detection framework with Principal Component Analysis PCA (PCA-LSTM-IDS). The PCA-LSTM-IDS is comprised of two layers: extracting layer which using PCA, and the anomaly BiLSTM IDS layer for classification. Through a comparison study, the effectiveness of the proposed approach for the task of detecting attacks is evaluated using the NSL-KDD dataset, which solves many of the inherent problems with the KDD'99 dataset. Experimental results indicate that the suggested framework, with an average accuracy of 99%, outperforms alternative approaches specially when using PCA with explained variance of 90%.