Kantharaju V, Suresh H, Niranjanamurthy M, Ansarullah Syed Immamul, Amin Farhan, Alabrah Amerah
Deparment of AI&ML, BMS Institute of Technology and Management (Affiliated to Visvesvaraya Technological University, Belagavi), Bengaluru, India.
Department of ISE, KNS Institute of Technology, Bengaluru, 560064, India.
Sci Rep. 2024 Dec 4;14(1):30275. doi: 10.1038/s41598-024-81535-3.
The Internet of Things (IoT) consist of a network of interconnected nodes constantly communicating, exchanging, and transferring data over various network protocols. Intrusion detection systems using deep learning are a common method used for providing security in IoT. However, traditional deep learning IDS systems do not accurately classify the attack and also require high computation time. Thus, to solve this issue, herein, we propose an advance Intrusion detection framework using Self-Attention Progressive Generative Adversarial Network (SAPGAN) framework for detecting security threats in IoT networks. In our proposed framework, at first, the IoT data are gathered. Then, the data are fed to pre-processing. In pre-processing, it restored the missing value using Local least squares. Then the preprocessing output is fed to feature selection. At feature selection, the optimum features are compiled using a modified War Strategy Optimization Algorithm (WSOA). Based upon the optimum features, the intruders were categorized into two categories named Anomaly and Normal using the proposed framework. Numerous attacks are assembled, including camera-based flood, DDoS, RTSP brute force, etc. We have compared our proposed framework using state of the art model and efficiency of 23.19%, 27.55%, and 18.35% higher accuracy and 14.46%, 26.76%, and 13.65% lower computational time compared to traditional models.
物联网（IoT）由一个相互连接的节点网络组成，这些节点通过各种网络协议不断地进行通信、交换和传输数据。使用深度学习的入侵检测系统是物联网中提供安全保障的常用方法。然而，传统的深度学习入侵检测系统不能准确地对攻击进行分类，并且需要较长的计算时间。因此，为了解决这个问题，在此我们提出一种先进的入侵检测框架，该框架使用自注意力渐进生成对抗网络（SAPGAN）来检测物联网网络中的安全威胁。在我们提出的框架中，首先收集物联网数据。然后，将数据输入到预处理阶段。在预处理中，使用局部最小二乘法恢复缺失值。接着将预处理输出输入到特征选择阶段。在特征选择阶段，使用改进的战争策略优化算法（WSOA）编译最优特征。基于这些最优特征，使用所提出的框架将入侵者分为异常和正常两类。收集了许多攻击类型，包括基于摄像头的洪水攻击、分布式拒绝服务（DDoS）、实时流协议（RTSP）暴力攻击等。我们将我们提出的框架与现有模型进行了比较，结果表明，与传统模型相比，我们的框架准确率分别提高了23.19%、27.55%和18.35%，计算时间分别降低了14.46%、26.76%和13.65%。
