An Improved Hybrid Sampling Model for Network Intrusion Detection Based on Data Imbalance

Network intrusion detection constitutes a pivotal element in safeguarding computer networks against malicious attacks and unauthorized access. With the widespread proliferation of the internet and the rapid evolution of information technology, network intrusions have become increasingly commonplace and intricate, underscoring the growing significance of network intrusion detection. In order to enhance the performance of network intrusion detection, we propose a sampling method that combines ADASYN with GMM, denoted as AGM. Firstly, the dataset undergoes preprocessing to mitigate noise, inconsistencies, and incompleteness issues inherent in the UNSW-NB15 dataset. Subsequently, sampling processing is applied to the dataset to mitigate the bias towards predicting majority classes, thereby improving prediction accuracy for minority classes. In conclusion, the amalgamation of CNN, BiLSTM, and Channel-Attention in the refined network architecture, CNN-BiLSTM-ATT, capitalizes on the distinctive advantages of each component. The classification outcomes of our experiments demonstrate the notable efficacy of the enhanced sampling technique and network structure for the task of network intrusion detection.