DBIS

Evaluation of cryptographic-based privacy-preserving machine learning libraries for URL phishing classification:
Phishing has been an issue since the earliest days of the birth of the Internet. Phishing techniques have improved over the years.
Up to this day, phishing accounts for one of the main sources of cybercrime. Phishing URLs can get distributed over social networks and messengers of any nature. While designed highly deceiving to manipulate a user into believing the legitimate status of these websites, the main goal of phishing is information theft for malicious motivations. Thus, it is inherently important to use the most recent and powerful tools, i.e., machine learning, to mitigate phishing attacks. Advancements in machine learning accelerate the adoption process for this approach with great potential for applicability in almost any sector. With the growing popularity of machine learning, respective privacy concerns gain more importance considering the current cyber state and privacy regulations. Nevertheless, privacy is not a central component of machine learning. Thus, while machine learning is not actively secured against privacy violating attacks with the goal of information theft, it is important to enhance the state of privacy of machine learning as early as possible. Aiming to examine the feasibility of applying privacy-enhancing technologies for private machine learning for URL phishing detection.