ML-Based Real-Time URL Inspection with Hardware Acceleration for Enhanced Web Security

With the proliferation of digital technology, a significant number of users leverage edge and IoT devices for web exploration. However, the inherent resource constraints of these systems often preclude the deployment of computationally intensive security mechanisms. Consequently, edge and IoT devices have become prime targets for cybercriminals. A critical issue arises when these devices encounter compromised websites capable of executing various cyberattacks, such as phishing, malvertising, and clickjacking. This paper proposes a machine learning-based approach for real-time URL inspection to provide immediate feedback on the legitimacy of accessed websites. We extract and analyze URL features to classify websites into categories such as benign, defacement, phishing, malware, and spam. We have developed a hardware-efficient neural network to perform this classification in real-time. Our solution can be integrated either as an extension to browser code or as a hardware module, thereby enhancing the security of edge and IoT devices during web browsing. The simulation results show an accuracy of nearly 98%, which is very favorable considering the real-time nature of the proposed model.