Metaheuristic-driven LSTM framework for advanced cyberthreat detection and response using a hybrid firefly-whale-grey...

Area of Science:

• Cybersecurity
• Artificial Intelligence
• Machine Learning

Background:

• Conventional signature-based Intrusion Detection Systems (IDSs) struggle against advanced cyber threats like APTs, zero-day exploits, and polymorphic malware.
• The dynamic nature of modern cyber threats necessitates more adaptive and intelligent detection mechanisms.

Purpose of the Study:

• To develop a novel, three-step hybrid optimization model for enhanced intrusion detection.
• To leverage the pattern-learning capabilities of Long Short-Term Memory (LSTM) networks combined with metaheuristic algorithms for improved accuracy and robustness.

Main Methods:

• A hybrid model integrating LSTM networks with three bio-inspired algorithms: Firefly Algorithm (FA) for feature selection, Whale Optimization Algorithm (WOA) for LSTM parameter tuning, and Grey Wolf Optimizer (GWO) for ensemble weight optimization.
• The framework was rigorously tested on the NSL-KDD and CICIDS2017 benchmark datasets using a protocol of 30 independent runs.

Main Results:

• The hybrid model achieved high detection accuracy: 98.62% on NSL-KDD (SD 0.21) and 99.1% on CICIDS2017 (SD 0.18).
• Achieved a low False Positive Rate (FPR) of 1.2% and a high Area Under the ROC Curve (AUC-ROC) of 0.992.
• Statistical significance was confirmed using Wilcoxon signed-rank and paired t-tests (p < 0.001).

Conclusions:

• The proposed hybrid optimization model demonstrates significant potential for near-real-time cyber threat detection.
• Further validation in real-world network environments is recommended to fully assess its practical applicability.