Towards a Privacy Preserving Framework for Mobility as a Service (MaaS)

Mobility-as-a-Service (MaaS) is reshaping urban transportation by integrating multiple travel modes into a cyberphysical ecosystem powered by artificial intelligence. However, growing privacy concerns and regulatory mandates, such as the GDPR Right to Be Forgotten (RTBF), pose significant challenges for data deletion and model compliance. To address this challenge, we propose a privacy-preserving machine unlearning (MU) framework tailored for MaaS using the Sharded, Isolated, Sliced, and Aggregated (SISA) approach. Our framework supports efficient removal of specific user data from trained ML models without requiring full retraining. Our evaluation shows that our framework can achieve high model utility, while effectively handling deletion requests (e.g., achieving a KL divergence of 0.0060-0.0068 and inference false positive rates of 17%-18%).