Network intrusion detection techniques based on improved multi-scale 1D convolutional neural networks

Abstract With the increasing number of network security threats, the requirements for network intrusion detection accuracy and real-time performance are getting higher and higher. The study creatively suggests a network intrusion detection method based on an enhanced multi-scale one-dimensional convolutional neural network to improve the precision and effectiveness of network intrusion detection. This technique extracts intrusion data features and improves the understanding of contextual information by enhancing the network. In addition, the model classification accuracy is improved by multi-head attention mechanism. The issue of data imbalance is addressed with a better synthetic minority class oversampling method. The experimental results indicated that the proposed network intrusion detection model outperformed the existing techniques in terms of detection precision, recall, F1 score, and other metrics. The model achieved 98.42% precision, 94.45% recall, and 96.40 F1 score on the attack samples. The proposed model was in low load condition with 22% CPU occupancy when processing 3000 sample data. Furthermore, the response time of the model was 0.52s. The outcomes reveal that the model has better detection performance and strong robustness, and has high application value in real-time response. The study provides a new and reliable basis for network intrusion detection techniques.