Optimising Intrusion Detection Through Long Short-Term Memory Model
Anjali Tharun, Amogh Srivathsa Kondapalli, S. Prabakeran · 2024
This research paper presents an innovative approach to enhancing Intrusion Detection Systems (IDS) using a Long Short-Term Memory (LSTM) model and transfer learning. The study begins with the loading of a pretrained LSTM model, followed by defining the architecture of the LSTM model. The weights from the pretrained model are transmitted to the new model, and the LSTM layer weights are frozen to preserve foundational knowledge. A classification layer is added to classify between normal network behavior and potentially malicious activities. The model is compiled and fine-tuned on an intrusion detection dataset to enhance its performance. The model's performance is continuously assessed and updated with fresh data to ensure accuracy. The system includes various modules such as Network Detection, Network Logging, IDS Protection, Attack Event Handler, Transfer Learning, and False Alarm Event Handler modules. The evaluation of the LSTM-based intrusion detection system on both training and testing datasets yielded promising results, demonstrating the model's robustness and generalization capabilities. The model exhibits high precision, recall, and F1-scores, along with consistent accuracy. A comparative analysis was conducted using a Decision Tree and Logistic Regression model alongside the proposed LSTM-based model, demonstrating the advantages of employing LSTMs for intrusion detection systems. The results affirm the robustness and reliability of the LSTM model in safeguarding network security against a wide range of cyber threats.