Adversarial Sample Generation using the Euclidean Jacobian-based Saliency Map Attack (EJSMA) and Classification for IEEE 802.11 using the Deep Deterministic Policy Gradient (DDPG)

Abstract

One of today's most promising developments is wireless networking, as it enables people across the globe to stay connected. As the wireless networks' transmission medium is open, there are potential issues in safeguarding the privacy of the information. Though several security protocols exist in the literature for the preservation of information, most cases fail with a simple spoof attack. So, intrusion detection systems are vital in wireless networks as they help in the identification of harmful traffic. One of the challenges that exist in wireless intrusion detection systems (WIDS) is finding a balance between accuracy and false alarm rate. The purpose of this study is to provide a practical classification scheme for newer forms of attack. The AWID dataset is used in the experiment, which proposes a feature selection strategy using a combination of Elastic Net and recursive feature elimination. The best feature subset is obtained with 22 features, and a deep deterministic policy gradient learning algorithm is then used to classify attacks based on those features. Samples are generated using the Euclidean Jacobian-based Saliency Map Attack (EJSMA) to evaluate classification outcomes using adversarial samples. The meta-analysis reveals improved results in terms of feature production (22 features), classification accuracy (98.75% for testing samples and 85.24% for adversarial samples), and false alarm rates (0.35%).