AN OPTIMIZED IRLS-SVM FRAMEWORK FOR IOT CYBERSECURITY: VULNERABILITY ASSESSMENT AND THREAT DETECTION

Abstract

The increasing deployment of Internet of Things (IoT) systems in smart cities, healthcare, and industrial environments has intensified cybersecurity challenges due to large-scale and evolving attack patterns. Traditional security mechanisms are often ineffective in detecting sophisticated and unknown threats. This paper presents an optimized machine learning–based framework for vulnerability assessment and threat detection in IoT networks. The proposed approach employs Z-score normalization for data preprocessing, L1-norm–based Mayfly Optimization (LNMFO) for selecting optimal features, and an Iteratively Reweighted Least Squares–based Support Vector Machine (IRLS-SVM) for robust classification. The optimization-driven feature selection reduces dimensionality and improves computational efficiency, while the IRLS mechanism enhances resilience to noise and outliers. Experimental evaluation on the BoTNeTIoT and HIKARI-2021 datasets demonstrates that the proposed framework achieves superior performance, with detection accuracy reaching 99%, outperforming existing intrusion detection methods. The results highlight the effectiveness of the proposed model for reliable IoT cybersecurity applications.