Network Intrusion Detection via Oversampling and Transformer-BiLSTM-MLP

This paper addresses the data imbalance problem in network intrusion detection. A novel network intrusion detection system(NIDS) is proposed, combining the adaptive synthetic sampling (ADASYN) algorithm with the transformer-bidirectional long short-term memory-multi-layer perceptron (TrBiM) model. Firstly, the ADASYN algorithm is employed to create a balanced dataset by addressing the imbalanced distribution of different types of data. Subsequently, the transformer model is utilized to capture the relationships between features while the bidirectional long short-term memory (BiLSTM) model is applied to analyze the temporal dependencies and sequential patterns in feature extraction. Finally, the MLP model maps the features extracted by bidirectional long short-term memory (BiLSTM) to the corresponding class labels for intrusion detection classification. Experimental results on the NSL_KDD dataset demonstrate the superiority of the proposed NIDS in terms of detecting minority class attacks and achieving a higher recall and f1-measure compared to other machine learning and deep learning algorithms.