An optimized LSTM-based deep learning model for anomaly network intrusion detection.

Area of Science:

• Computer Science
• Cybersecurity
• Artificial Intelligence

Background:

• Increasing network connectivity necessitates robust network security and cyberattack defense.
• Intrusion Detection Systems (IDS) are crucial for identifying network breaches, but traditional methods often have high false alarm rates.
• Machine learning and deep learning offer promising solutions for enhancing IDS efficacy.

Purpose of the Study:

• To propose an optimized Long Short-Term Memory (LSTM) model for accurate network traffic anomaly detection.
• To mitigate the high false alarm rates associated with conventional intrusion detection systems.
• To evaluate the effectiveness of deep learning methodologies in cybersecurity applications.

Main Methods:

• An optimized Long Short-Term Memory (LSTM) network was developed for anomaly identification in network traffic.
• Three optimization algorithms—Particle Swarm Optimization (PSO), JAYA, and Salp Swarm Algorithm (SSA)—were employed to tune LSTM hyperparameters.
• The proposed models were evaluated using the NSL KDD, CICIDS, and BoT-IoT datasets.

Main Results:

• A comparative analysis was performed on PSO-LSTMIDS, JAYA-LSTMIDS, and SSA-LSTMIDS.
• The Salp Swarm Algorithm-optimized LSTM (SSA-LSTMIDS) model achieved superior performance across all tested datasets.
• Performance was assessed using metrics including Accuracy, Precision, Recall, F-score, True Positive Rate (TPR), False Positive Rate (FPR), and Receiver Operating Characteristic (ROC) curves.

Conclusions:

• Optimized deep learning models, particularly SSA-LSTM, significantly improve network intrusion detection capabilities.
• The proposed SSA-LSTM model offers a viable and effective solution for reducing false alarms in intrusion detection systems.
• This research highlights the potential of advanced optimization algorithms in enhancing deep learning-based cybersecurity solutions.