A convolutional neural network-enhanced attack detection framework with explainable artificial intelligence for internet of things-based metaverse security

Abstract

Metaverse integration is deeply reliant on the Internet of Things (IoT), which enables seamless data exchange between the physical and virtual worlds. By connecting sensors, wearables, and smart devices to Metaverse environments, IoT enhances interactions and immersion, essential for a fully integrated, data-driven Metaverse. However, IoT devices, due to their simple hardware and Internet access, are vulnerable to cyberattacks, making security a critical challenge for ensuring a safe infrastructure. Traditional methods struggle to combat evolving threats, requiring adaptive artificial intelligence (AI) powered solutions. This work presents a hybrid framework combining convolutional neural networks (CNN) with machine learning (ML) classifiers, categorical boosting (CatBoost), and adaptive boosting (AdaBoost), enhanced by metaheuristic optimization for better performance. A two-tier architecture is proposed for handling complex data to detect and identify attacks within IoT networks. A comparative analysis of a real-world IoT dataset demonstrates the framework’s effectiveness in detecting threats (binary classification) and identifying malicious attacks (multi-class classification). A total of six experiments are conducted using publicly available data with the best models achieving .993 accuracy for binary classifications and .877 for multi-class. Explainable AI techniques further provided insights into model decisions, guiding future data collection and system security.