An ensemble federated learning framework for privacy-by-design mobility behaviour inference in smart cities

Abstract

Inferring the travel behaviour of users in their GPS trajectories, while protecting their privacy is a significant issue for smart and sustainable cities. To address this challenge, we use Federated Learning (FL), a privacy-preserving machine learning technique that aims at collaboratively training a robust global model by accessing users’ locally trained models, but not their data. Specifically, we design a novel eNsemble federATed leArning for mobiLity InfErence (NATALIE) framework. The ensemble method combines the outputs from different DNN models learned via FL and shows an accuracy that surpasses comparable models reported in the literature. Extensive benchmarking experiments on open-access MTL Trajét and GeoLife GPS datasets demonstrate that the proposed inference model can achieve comparable accuracy in the identification of mode of travel without compromising privacy. The evaluation of the proposed model against non-i.i.d. data at varying sample sizes and different worker numbers shows improved performance. Findings are expected to contribute to the advancement of the transportation sector in smart and sustainable cities.