A Novel Deep Learning Approach for Detecting Types of Attacks in the NSL-KDD Dataset

Abstract

 The growing prevalence of Internet intrusions poses significant threats to the security, privacy, and reliability of systems and networks. Denial-of-service (DoS) attacks are a cause for concern as they aim to disrupt access to network resources, posing major risks. Traditional intrusion detection systems (IDS) face challenges in detecting attacks because of the evolving nature of these attacks. Therefore, advanced techniques are necessary to accomplish accurate and timely detection. This study introduces a novel approach that combines Deep learning techniques, specifically the CNN algorithm, with Principal Component Analysis (PCA) and Singular Value Decomposition (SVD) for the purpose of feature selection. The effectiveness and efficiency of our method are shown by rigorous testing on DDoS datasets. We present a novel Fast Hyper Deep Learning Model that attains a remarkable accuracy of 99%, along with perfect recall and F1-measurement scores of 100%. This model surpasses existing methodologies by a significant margin. The NSL-KDD data set allows for achieving a level of precision of100%.