A novel feature selection algorithm for IoT networks intrusion detection system based on parallel CNN-LSTM model

Abstract

As the Internet of Things networks expand globally, it is becoming increasingly important to protect against threats. one of the main reasons for the high number of false positives and low detection rates is the presence of redundant and irrelevant features. To address this problem, we propose a binary chimpanzee optimization algorithm for the feature selection process. This paper presents accurate network-based intrusion detection network, named parallel convolutional neural network long and short-term memory network branch, which has two branches. The input vector of the network is permuted in a 3-dimention space. This allows the model to extract highly discriminative features using a small number of layers. On the second branch, we used long and short-term memory network in parallel. The efficacy of the proposed deep model has been evaluated using three benchmark internet of things intrusion detection datasets, namely ToN-IoT, UNSW-NB15, and IoTID20 datasets. The experimental results demonstrated that the proposed binary chimpanzee optimization approach reduces about 60% of features, and the effectiveness of the proposed model was demonstrated by experimental results showing a high detection rate, high accuracy, and a relatively low false positive rate, which are measured as 99.54%, 99.56%, and 0.024 in the ToN-IoT and 99.79%, 99.78%, and 0.0032 in UNSW-NB15 and 100%, 100%, and zero in IoTID20 datasets, respectively.