Scholarlink

<p>Traffic flow prediction models must be reliable, scalable, and privacy-preserving as urbanization and vehicle density increase. Centralised machine learning traffic forecasting methods store sensitive city-level data on servers, presenting privacy and logistical difficulties. Multiple cities may use Federated Learning to develop a worldwide traffic prediction model without sharing raw traffic data to solve these issues. Local deep learning models use GPS, weather, and road network data from each city. Federated averaging techniques combine model updates collected securely in a central aggregator to produce a global model. After redistribution to cities, this model allows data sovereignty and learning. It protects privacy, avoids network drift and participant dropout, and distributes city-wide data heterogeneously. With extensive testing on multi-city traffic datasets, the federated model matches centralized techniques in prediction accuracy while protecting data. Scalable cross-city traffic forecasts and intelligent transportation systems that follow contemporary data governance requirements are possible with this research.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Privacy-Preserving Federated Learning Framework for Collaborative Urban Traffic Flow and Disaster Management Across Cities

  • 期刊论文

  • 发表日期：

    2026年5月9日

查看全文
  • Chopparapu Gowthami,
  • S. Kavitha

摘要

Traffic flow prediction models must be reliable, scalable, and privacy-preserving as urbanization and vehicle density increase. Centralised machine learning traffic forecasting methods store sensitive city-level data on servers, presenting privacy and logistical difficulties. Multiple cities may use Federated Learning to develop a worldwide traffic prediction model without sharing raw traffic data to solve these issues. Local deep learning models use GPS, weather, and road network data from each city. Federated averaging techniques combine model updates collected securely in a central aggregator to produce a global model. After redistribution to cities, this model allows data sovereignty and learning. It protects privacy, avoids network drift and participant dropout, and distributes city-wide data heterogeneously. With extensive testing on multi-city traffic datasets, the federated model matches centralized techniques in prediction accuracy while protecting data. Scalable cross-city traffic forecasts and intelligent transportation systems that follow contemporary data governance requirements are possible with this research.