Enhancing IoT intrusion detection with genetic algorithm-optimized convolutional neural networks

The increasing complexity and volume of cyberattacks necessitate the development of advanced Intrusion Detection Systems that are capable of efficient and scalable threat detection. As network infrastructures expand exponentially and the proliferation of Internet of Things (IoT) devices, traditional intrusion detection systems face significant challenges in adapting to evolving attack patterns. This paper introduces a novel intrusion detection framework, leveraging Neural Architecture Search to optimize the design of a convolutional neural network (CNN). Specifically, it proposes a system that integrates CNN with genetic algorithm for both synthesizing the neural architecture and systematically optimizing associated hyperparameters. This approach improves model performance by automating the search for the most effective neural architecture. The proposed approach was evaluated on UNSW-NB15, TON_IoT, and CICIoT2023 datasets. Experimental results demonstrate the efficacy and scalability of the method, with 85.7%, 99.93%, and 90.13% classification accuracies, respectively. Moreover, the high classification results on the TON_IoT and CICIoT2023 datasets confirm the approach efficiency and suitability for real-time applications, addressing the practical challenges of securing IoT networks.