Enhancing IoT Intrusion Detection with Greylag Goose Optimization and Extreme Learning Machine: A Data-Driven Study on IoT23

The need for effective intrusion detection schemes has been heightened by the exponential development of internet of things (IoT) devices in today's smart environments, which greatly increases the likelihood of complex cyberattacks. Utilizing the IoT23 dataset, this research introduces a new, lightweight intrusion detection system (IDS) architecture designed for IoT networks. The proposed workflow consists of three stages: fast classification, clever data preparation, and metaheuristic feature selection. The preprocessing module addresses important challenges in raw network flow data, including data cleansing, categorical encoding, normalization, binary and multi-class label translation, and the control of class imbalance using synthetic oversampling methods. To reduce dimensionality and improve classification efficiency, we propose a greylag goose optimization (GGO) technique for feature selection that identifies a minimal yet highly relevant set of features. The GGO algorithm models its actions after the natural migration patterns of greylag geese to find the sweet spot between exploring and exploiting the feature space. The selected features are then inputted into an extreme learning machine (ELM), a single-layer feedforward neural network known for its lightning-fast training time and exceptional accuracy. Experimental findings show a multi-class accuracy of 92.56% and a binary classification accuracy of 96.85%, outperforming more traditional models like SVM, KNN, and random forest. The suggested IDS architecture is ideal for deployment in resource-constrained IoT applications as it improves detection accuracy while reducing computational overhead. The integration of evolving intelligence with lightweight learning techniques enables the proactive management of IoT security and the identification of threats in real-time.