Hybrid Deep Neural Networks for Advanced Intrusion Detection in Cybersecurity

Abstract

The rapid growth of cloud services and IoT devices has significantly increased network traffic, leading to a surge in sophisticated cyberattacks. Traditional intrusion detection systems (IDS) often fail to identify evolving and zero-day threats. To address these challenges, this study proposes a hybrid model that integrates Convolutional Neural Networks (CNN) for automated feature extraction with Random Forests (RF) for robust classification. Using the NSL-KDD dataset, the hybrid CNN-RF model achieved superior performance compared to standalone deep learning and machine learning approaches, attaining an accuracy of 99.2% with a false positive rate of only 0.5%. The model also demonstrated strong generalization in detecting zero-day attacks and provided interpretable feature importance insights. While slightly more computationally intensive in training, the prediction efficiency remained competitive. These findings highlight the potential of combining deep learning and ensemble methods to design reliable, scalable, and interpretable IDS solutions. Future work will explore real-time deployment, advanced architectures such as RNNs and transformers, and federated learning for enhanced privacy-preserving security.