Technology for Balancing Privacy and Accuracy in Mobility

2024 JST AIP Challenge:

This research aims to develop an innovative taxi-demand prediction system that prioritizes customer privacy while maintaining high prediction accuracy. The system leverages federated learning, a cutting-edge machine learning technique, enabling multiple taxi service providers to collaboratively train a predictive model without the need to share sensitive trajectory data. By decentralizing the training process and keeping data localized, this approach addresses significant privacy concerns and mitigates the risks associated with data breaches and misuse.

Federated learning facilitates the aggregation of knowledge from diverse data sources while preserving data privacy, thus overcoming the technical challenges inherent in conventional data-driven models. This decentralized approach ensures that the predictive model benefits from a wide range of data inputs, leading to improved accuracy and robustness in predicting taxi demand patterns.

This project encapsulates a pioneering effort to balance the utility of spatio-temporal data with stringent privacy requirements. By integrating advanced privacy-preserving techniques, the research aims to enhance the reliability and acceptance of data-driven solutions in urban transportation systems. The successful implementation of this system has the potential to revolutionize the way taxi services operate, leading to more efficient resource allocation, reduced wait times for passengers, and overall improved service quality.