An Approach to Classify Intrusion in IoT System Using Neural Architecture Search

By the end of this decade, approximately 25 to 50 billion Wi-Fi-enabled Internet of Things (IoT) devices will be available worldwide. These devices will have several vulnerabilities. So, a Network Intrusion Detection System (NIDS) is much needed; it will enhance the security of these devices. Several researchers have already proposed many NIDS. In this paper, on top of Deep Neural Networks (DNN) and Long Short-Term Memory (LSTM), we are using Random Neural Architecture Search (RNAS). Using the Aegean Wi-Fi Intrusion Detection dataset with NAS significantly reduced the manual tuning time and increased detection accuracy compared to the previous deep learning models. In our work, we classified the models into binary and multi-class categories. For the DNN with NAS in binary classification, we achieved an accuracy of 97.81%. In multi-class classification, the accuracy was 95%. For the LSTM model with NAS, the accuracy in binary classification was 97%, while in multi-class classification, it was 97.21%.