A BiLSTM Approach to Enhance the Accuracy of SQL Injection Attack Detection

This paper proposes an innovative method for detecting SQL injection attacks by leveraging the temporal modeling capabilities of Bidirectional Long Short-Term Memory (Bi-LSTM) neural networks. Our approach captures the sequential relationships between traffic packets before and after an intrusion by utilizing a labeled NetFlow V5 dataset. By introducing timestamp information and reprocessing the dataset, the proposed method effectively learns the temporal characteristics of attacks. Comparative experiments demonstrate the superiority of the Bi-LSTM model over unidirectional LSTM and fully connected neural networks, maintaining a false positive rate below 0.22% on the public test set. Furthermore, by incorporating additional injection traffic features, the model's adaptability and effectiveness in real-world attack scenarios are enhanced. The proposed Bi-LSTM based method provides an accurate and efficient solution for detecting SQL injection attacks by using NetFlow data, highlighting the importance of temporal modeling in network intrusion detection.