Hierarchical Deep Learning with Nature-Inspired Optimization for Robust Network Intrusion Detection

Abstract

This study proposes a Swarm-Based Multi-Layer Intrusion Detection System (SML-IDS) that combines hierarchical deep learning and swarm intelligence-based feature optimization to enhance network security. Our framework integrates Convolutional Neural Networks (CNNs) for packet-level anomaly detection, Long Short-Term Memory (LSTM) networks for session-level behavior analysis, and fuzzy logic-based context analyzers to minimize false alarms. Feature selection is optimized through bio-inspired Elephant Herding Optimization (EHO), improving classification accuracy while reducing computational overhead. Evaluation on CIC-IDS2017 dataset shows that SML-IDS outperforms conventional IDS models, achieving superior detection accuracy, false positive rate reduction, and real-time feasibility.