Enhanced Federated Learning Architecture for Secure Location Privacy

Location based services (LBS) have utilized mobile computing applications to deliver personalized, real time services including navigation, activity monitoring, and contextual recommendations.However, the centralized system for gathering and processing sensitive location information is highly time consuming and problematic, primarily due to the risk of re-identification attacks, including Point of Interest, Probabilistic Inter Point of Interest Transition Attack, and All Points Attack.This paper presents a privacy preserving framework that leverages decentralized learning to optimally balance location data utility with robust privacy protection via Paillier Homomorphic Encryption as a trajectory-level privacy mechanism, integrated with federated learning, a partial encryption technique.The present work comprises three privacy methods: Geo Indistinguishability, Geometric Triangulation, and Homomorphic Encryption, to select the optimal Location Privacy Protection Mechanisms for each user's real mobility in the federated learning environment.Compared to conventional centralized or rule based models, this model integrates the Long Short Term Memory to learn the user's spatial and temporal mobility.The proposed framework, using the Geolife dataset collected by Microsoft Research Asia, was applied to real world trajectories of 180 users to simulate realistic mobility patterns.Each user functioned as an independent federated client, contributing local model updates aggregated centrally every 24 hours.Experiments were run on Google Colab using Python based frameworks such as 'phe' for PHE, TensorFlow for deep learning, and Flower for federated learning.The simulation results show that the proposed framework improves privacy scores by 52% against re-identification attacks and increases the usefulness of 20% of the original data.The proposed method offers a scalable and secure solution for preserving user privacy while minimizing data communication, providing a viable approach for real world mobile LBS applications.