ML filter for USB threat detection

This thesis project aims to investigate the feasibility of using an AI/ML-based approach to mitigate threats. In collaboration with Saab, this project will in- vestigate the efficacy of using these algorithms to monitor USB traffic and, from this, determine if a USB device operates as it should. My stakeholder specifically has an interest in determining that confidential files are not ex- tracted to, for example, a modified USB keyboard. Tasks for the project include studying preexisting solutions and relevant algorithms, gathering and pro- cessing data, evaluating algorithms, and implementing and training a ma- chine learning model. The implementation will be evaluated using synthetic scenarios modeled on realistic attack patterns that involve both harmless and malicious USB traffic. This thesis has the potential to be a contribution to USB security by providing a reproducible solution to evolving threats.