Hybrid Ensemble Deep Neural Network for Intrusion Detection (HEDNN-ID)

Abstract

As network traffic becomes increasingly complex today, the variety of cyber attacks is also increasing, and the need to prevent these attacks in real-time is emerging; network intrusion detection systems (NIDSs) are essential. In contrast, traditional IDS methodologies, including rule-based and statistical approaches, often face challenges in maintaining their effectiveness as they struggle to keep pace with the rapid development of Large-Scale, dynamic traffic and the evolving behavior of attackers. Although CNN-LSTM and transformer-based frameworks improve detection accuracy using state-of-the-art deep learning models, the challenges of addressing spatial-temporal dependencies, class imbalance, and scalability remain to be addressed. Such challenges necessitate a robust, fast, and scalable framework for network-level intrusion detection. This research proposes a Hybrid Ensemble Deep Neural Network model termed HEDNN-ID to address these limitations. This model explicitly integrates attention mechanisms to focus on significant data, Long Short-Term Memory (LSTM) to learn temporal dependencies, Convolutional Neural Networks (CNN) to extract spatial features, and ensemble learning to enhance generalization and resilience. HEDNN-ID compared favorably with leading models, achieving 98.68% accuracy, 97.80% precision, 97.50% recall, and 97.65% F1-score on the UNSW-NB15 dataset. The proposed framework effectively addresses the limitations of existing approaches and supports scalability for practical use cases in safe intrusion detection. HEDNN-ID can adapt to various attack scenarios and enhance detection reliability, which represents a significant step forward in modern cybersecurity. The research provides a foundation for developing impenetrable, scalable IDS frameworks.