Metaheuristic Optimization Algorithms for Deep Learning Model Design in Secure Internet of Things Environment

Abstract

The Internet of Things (IoT) has enabled smart systems, but it has also increased vulnerabilities to cyber threats, including botnet attacks. To address these security challenges, this study proposes a hybrid system that combines metaheuristic and machine learning. To tune hyperparameters of a hybrid neural network based on Convolutional Neural Networks and Semi-Recurrent Neural Networks (CNN-QRNN), the Chaotic Butterfly Optimization Algorithm (CBOA) is used. A new metaheuristic algorithm, Self-Adaptive Enhanced Harris Hawks Optimization (SAEHO), as well as a self-upgraded cat and mouse optimizer (SU-CMO), are introduced and evaluated in order to enhance model effectiveness. Based on experiments conducted on the N-BaIoT dataset, it was determined that the proposed models significantly outperformed conventionalclassifiers in key performance metrics, including accuracy, the Matthews Correlation Coefficient (MCC), theRand Index, and the F-Measure. Particularly notable improvements were observed in reducing false-positiverates and enhancing anomaly detection sensitivity. The HMMLB-BND method substantially improvesdetection performance in diverse IoT environments, offering a robust, efficient, and scalable solution suitablefor real-time deployment in resource-constrained systems.