An Improved Lightweight CNN_BiLSTM Model for Network Intrusion Detection Using the UNSW-NB15 Dataset

The attack behavior in the network poses a threat to security and privacy issues. In order to improve the effectiveness of IoT network attack detection and compensate for the shortcomings of traditional machine learning methods in detection capabilities, a network intrusion detection method based on lightweight convolutional neural networks (CNN) and bidirectional long short-term memory networks (BiLSTM) is proposed. This method preprocesses data through normalization techniques, extracts data features from both spatial and temporal dimensions using CNN and BiLSTM models, and inputs the extracted spatiotemporal feature vectors into the CNN BiLSTM upper classifier for training, achieving intrusion detection. The proposed model was validated on the dataset UNSW-NB15, and the results showed that the binary and multi classification accuracy of the CNN_BiLSTM model was around 93%. Compared with other methods, the proposed method significantly improves the accuracy of network intrusion detection, and also demonstrates better performance in detection accuracy and false positive rate.