Applying Modified Convolutional Neural Networks for Detecting Intrusion
Kanin Siritharagul, Sunantha Sodsee · 2023
The rapidly evolving landscape of cyber threats necessitates more adaptive and efficient intrusion detection systems (IDS). To meet this need, we propose a novel approach that shifts IDS from traditional machine learning techniques to deep learning, specifically leveraging Convolutional Neural Networks (CNN). Our method innovatively transforms network flow data from the NSL-KDD dataset into images. This allows for the application of CNN to analyze these images and classify types of intrusion attacks. Furthermore, we utilize the pre-trained VGG16 model for transfer learning, enhancing the initial training efficiency. We then implement incremental learning (IL) to continuously adapt to and learn from new types of attacks, addressing the dynamic nature of cyber threats. The potential enhancements achievable through hyperparameter optimization (HPO) are also explored. The results demonstrate an accuracy of 82.57% and a precision of 86.94%, emphasizing the effectiveness of our approach in advancing the development of adaptive IDS within the continuously evolving digital environment.