3DLBS-BCHO: a three-dimensional deep learning approach based on branch splitter and binary chimp optimization for intrusion detection in IoT

3DLBS-BCHO: a three-dimensional deep learning approach based on branch splitter and binary chimp optimization for intrusion detection in IoT

연구 분야: Networking

학회: Cluster Computing

The rapid expansion of the Internet of Things networks has increased the difficulty of maintaining network security and identifying cyberattacks within these networks. This paper proposed a three-dimensional Deep Learning approach based on Branch Splitter (3DLBS). The proposed 3DLBS transforms one-dimensional data into two-dimensional data using shape and fill techniques and permute techniques for transforming to 3D data. The data is then sent into two CNN and LSTM branches to detect attacks. This paper introduces the Binary Chimp Optimization algorithm (BCHO) for feature selection, resulting in improved accuracy and speed of intrusion detection. The algorithm was applied to the ToN-IoT and UNSW-NB15 datasets to extract essential features, reducing the number of features by an average of 52%. Reduced features sent to the DLBS to increase the speed and accuracy of intrusion detection (3DLBS-BCHO). Experiments showed high accuracies of 99.54% on the ToN-IoT dataset and 99.93% on the UNSW-NB15 dataset.