In today’s world, electronics and networked systems, such as IoT devices, embedded platforms and smart environments, are increasingly popular and widespread. As a result, these systems become more exposed to cyber threats. The malicious URL is also one of the most widespread yet perilous vectors of cyberattack, as it is widely used in phishing, malware distribution, and command-and-control communication. The security of these electronic systems necessitates real-time, lightweight and intelligent detection techniques that must be efficient in resource-constrained environments. In order to meet this requirement, we propose SwiftURL, a lightweight deep learning model to detect malicious URLs that can be specifically deployed in modern electronic environments. SwiftURL leverages knowledge distillation from a transformer-based ELECTRA-Small teacher model, transferring detection capability into a smaller and faster student model while maintaining high performance. Experimental results on a public Kaggle dataset of malicious URLs demonstrate that SwiftURL achieves an accuracy of 94.38%, reduces computational overhead by 35%, and accelerates training time by 15%. These findings highlight SwiftURL’s effectiveness as a practical solution for enhancing cybersecurity in electronic and networked systems through efficient, on-device URL threat detection.
Lim et al. (Mon,) studied this question.