Self-Adaptive Intrusion Detection Framework for IoT Networks Using Addax Optimization and Echo State Networks

Abstract

The rapid rise of the Internet of Things (IoT) has rendered its networks vulnerable to sophisticated attacks, necessitating the use of advanced intrusion detection techniques. This paper presents an innovative approach to IoT intrusion detection that combines advanced preprocessing approaches, hyperparameter optimization, and effective classification algorithms. The preprocessing pipeline employs Deep Denoising Autoencoders (DDA) to remove noise from IoT sources, Kalman Filters (KF) to smooth temporal data, and Dynamic Range Compression (DRC) to standardize sensor readings. The framework dynamically adjusts the hyperparameters and integrates the Self-Adaptive Addax Optimization Algorithm (AOA) to select the optimum model architecture for various attack situations, resulting in optimal model performance. This involves integrating ESNs, GRUs, and GANs into a single network architecture through the use of a attack categorization model. Its primary impact is to strengthen the system as a whole while maintaining precise detection even at intricate levels and with a high degree of pattern flexibility. The system achieves outstanding performance with 98.45% accuracy, 97.92% precision, 98.65% recall, and an F1 score of 98.27% when evaluated on the NSL-KDD dataset. Scalable, adaptable, and effective intrusion detection systems are provided by the integrated solution, which has resolved the security challenges posed by dynamic IoT networks