Designing A Network Intrusion Detection System Using Advanced Bi-LSTM Based Model for Identifying and Classifying Intrusions

By enabling the control and administration of the entire network from a single location, a software-defined network, or SDN, was created to make network administration easier. Today’s data center network infrastructure frequently uses SDN, but emerging threats like Distributed Denial-of-Service (DDoS), online attacks, and the User to Root (U2R) attack pose serious problems that could prevent SDNs from being widely adopted. SDN controllers find intruders appealing because they are worthwhile targets. An attacker may take control of an SDN controller and use it to route traffic anyway they see fit, which would have disastrous effects on the network as a whole. The quality of the training datasets affects how well the detection models work, even though the combined vision of SDN and deep learning techniques creates new opportunities for the security of IDS deployment. Most of the studies ignored the effects of data redundancy and an unbalanced dataset, despite the fact that deep learning for NIDSs has recently demonstrated encouraging outcomes for a variety of concerns. This could therefore have a negative impact on the anomaly detection system’s robustness, leading to less-than-ideal model performance. We imported all necessary libraries, including the UNSW-NB15 training dataset, and used the proposed model in this investigation. We are combining the Random Forest model with the BiLSTM model, a deep learning method. The combined output is then analyzed to determine the accuracy, precision, recall, and F1-score.