A Modified Transformer Neural Network (MTNN) for Robust Intrusion Detection in IoT Networks
Syed Wahaj Ahmed, Fabio Kientz, Rasha Kashef
Abstract
The growth of the Internet of Things (IoT) in various industries has been unprecedented over the past few decades. However, IoT devices are prone to malicious network entities (i.e., attacks) such as data theft, phishing, spoofing, and denial of service attacks (DDoS attacks). These can result in additional cyber security risks, such as ransomware attacks and significant data breaches, which can cost firms a lot of money and time to repair. Thus, there is a great demand and need for building robust Intrusion detection systems (IDS) for real-time identification and recognition of these attacks. With the advances in neural networks, various models have been proposed. However, most traditional models lack the detection of diverse attack types due to their limited adaptability. Thus, in this paper, we propose a network Intrusion Detection System based on the attention mechanism of transformer neural networks, namely, the MTNN model. For performance evaluation, a traditional LSTM and RNN are also implemented and compared to the proposed model using the ToN_IoT dataset. Experimental results show that the MTNN model has achieved an improvement of up to 57%, 33%, 70%, and 63% for accuracy, precision, recall, and F-score, respectively.