Federated Reinforcement Learning for Consumers Privacy Protection in Mobility-as-a-Service

Abstract

Mobility-as-a-Service (MaaS) offers multi-modal transport modes in a single service platform, which requires tremendous data and software support. Among various types of data, consumers' data is vulnerable to the communication channel as it must be transmitted from the consumer end to the MaaS. Consumers put a high priority on the privacy of their data in selecting a service. This motivates the need for a secure information management system for MaaS to protect consumers' information from leakage. In this paper, we propose a federated reinforcement learning (FRL) approach for the information exchange intensive multi-modal journey planning process. The FRL approach protects the information from malicious information thieves by federating the global model training to a local one without sensitive information exchange while maintaining the same solution quality of enhancing MaaS profit and consumer satisfaction. We perform experiments on a test case based on New York City data. The results demonstrate that the FRL approach is effective in the MaaS multimodal journey planning process. Compared to the baseline approaches, consumer satisfaction and MaaS profit increase by about 12% and 74%, respectively. This pilot study not only provides privacy protection insight into the MaaS multi-modal journey planning but also other privacy-concern applications.