Copyright © 2025 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{189540,
        author = {GIRIJA GM and DURGA NAYANA NIKITHA G and PRIYANKA B and DESHANUR SWETHA and SURESH K},
        title = {Deep Learning Based Intrusion Detection System},
        journal = {International Journal of Innovative Research in Technology},
        year = {2025},
        volume = {12},
        number = {7},
        pages = {5843-5848},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=189540},
        abstract = {Historically, intrusion detection systems (IDS) have played a vital role in safeguarding network infrastructures against cyberattacks. Early IDS solutions predominantly relied on signature-based detection techniques, which were effective for identifying known attack patterns but exhibited significant limitations when confronted with zero-day attacks, polymorphic malware, and adaptive or moving-target threats. To mitigate these shortcomings, anomaly-based detection methods were introduced to identify deviations from normal network behavior. Although anomaly-based IDS improved the detection of previously unseen attacks, they often suffered from high false-positive rates and limited contextual awareness.
With the increasing complexity and scale of modern cyber threats, deep learning (DL) techniques have emerged as a powerful paradigm for intrusion detection. Deep learning-based IDS (DL-IDS) are capable of automatically learning hierarchical and high-level representations from high-dimensional network traffic data, leveraging advances in computational resources and neural network architectures.
In this paper, we propose a robust hybrid DL-IDS framework that integrates both signature-based and anomaly-based detection mechanisms, further strengthened through an ensemble of deep learning models. Unlike traditional approaches that rely on classical machine learning algorithms, the proposed system exclusively employs deep neural networks. Specifically, convolutional neural networks (CNNs) are utilized to capture spatial feature representations, recurrent neural networks (RNNs) to model temporal dependencies in sequential traffic data, and Transformer models to extract context-aware features using attention mechanisms. Each model is independently trained on benchmark intrusion detection datasets, including NSL-KDD and CICIDS2017, following comprehensive data preprocessing, normalization, and feature encoding procedures.
To enhance detection robustness and reduce model bias, a majority-voting fusion layer is employed to aggregate predictions from the individual deep learning models. This ensemble strategy improves stability and resilience across diverse and evolving attack scenarios.
Experimental evaluations demonstrate that the proposed ensemble DL-IDS achieves detection accuracy exceeding 96%, with precision, recall, and F1-score consistently surpassing 94% across all attack categories. The attention-based Transformer component significantly enhances generalization to previously unseen threats, while the CNN and RNN branches effectively capture complementary spatial and temporal characteristics of network traffic. Furthermore, the modular and scalable architecture of the proposed system makes it well suited for enterprise-scale networks, cloud-based environments, and real-time intrusion monitoring, offering low-latency processing and the ability to handle high-volume traffic streams.
In addition, the proposed framework effectively addresses key challenges in intrusion detection, including class imbalance, feature redundancy, and the high dimensionality of network traffic data. Overall, this work demonstrates the practical viability and effectiveness of modern deep learning techniques in intrusion detection. By combining hybrid detection strategies with ensemble deep learning architectures, the proposed system achieves high accuracy, robustness, and operational reliability for real-world cybersecurity applications.},
        keywords = {Intrusion Detection System, Deep Learning, CNN, RNN, Transformer Model, Network Security, NSL-KDD, CICIDS2017, Ensemble Architecture, Cybersecurity},
        month = {December},
        }