Deep Learning Approaches for Intelligent Intrusion Detection Systems: Bridging Computer Science and Cybersecurity

Abstract Intrusion Detection Systems (IDS) are a cornerstone of modern cybersecurity, designed to safeguard networks from increasingly sophisticated attacks. Traditional machine learning–based IDS approaches often suffer from limited feature representation, high false alarm rates, and difficulty in adapting to evolving threat landscapes. To address these limitations, this study introduces DeepIDS-Net, a deep learning–driven intrusion detection framework that integrates convolutional and recurrent neural architectures to capture both spatial and temporal dependencies in network traffic. The proposed model was trained and evaluated on the widely adopted NSL-KDD dataset, which contains labeled records of normal and malicious activities across multiple attack categories. Preprocessing steps included data normalization, categorical encoding of protocol and service attributes, and feature scaling to ensure stable training. Experimental evaluation demonstrated that DeepIDS-Net achieved an accuracy of 98.4%, a precision of 97.9%, a recall of 98.1%, and an F1-score of 98.0%, significantly outperforming baseline models such as Random Forest, SVM, and standard deep feedforward networks. Key contributions include: (1) a hybrid deep architecture optimized for intrusion detection, (2) a systematic preprocessing pipeline enhancing learning efficiency, and (3) empirical evidence of reduced false positives while maintaining high detection sensitivity. The results highlight the potential of deep learning to transform IDS into adaptive, intelligent defense mechanisms. This work not only bridges computer science and cybersecurity but also provides a scalable pathway for real-world deployment of AI-enhanced IDS. Future research will extend this approach to large-scale, real-time streaming data for next-generation cyber defense.