Scholarlink

<p> <!--Query ID="Q1" Text="Please check and confirm if the authors and their respective affiliations have been correctly identified. Amend if necessary." Resolved="yes"-->This paper presents a privacy-preserving federated learning framework for urban traffic flow prediction using a Cheon-Kim-Kim-Song (CKKS) homomorphic encryption scheme. The proposed hierarchical system model consists of vehicles, roadside units (RSUs), and a central server, enabling collaborative traffic data analysis while maintaining data privacy. The performance of the federated learning system was evaluated with and without the CKKS encryption. In the CKKS-enabled configuration, local models initially exhibit higher errors, with mean squared errors (MSE) ranging from 0.1375 to 0.1621 and negative R² values. However, after ten training rounds, the models show significant improvements, with MSE dropping to 0.0007–0.0008 and R² values exceeding 0.9820, indicating effective learning of traffic patterns. In contrast, the non-encrypted federated learning system demonstrates strong performance from the outset, with MSE values around 0.0007–0.0008 and R² values of 0.98, further improving to an MSE of 0.0003 and R² approaching 0.993 by the end of training. While the non-encrypted system achieves high accuracy, it compromises data privacy. This study highlights the importance of balancing the predictive performance and data protection in privacy-sensitive applications. The proposed framework has practical implications for urban traffic management, enabling real-time data analysis without exposing personal identifiable information. Future research directions include exploring alternative encryption methods, optimizing architectures for faster convergence, and applying the framework to other domains such as smart cities and autonomous vehicles.</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 hierarchical federated learning approach for secure urban mobility prediction

  • 期刊论文

  • 发表日期：

    2025年7月4日

查看全文
  • Vankamamidi S. Naresh,
  • D. Ayaaappa

摘要

This paper presents a privacy-preserving federated learning framework for urban traffic flow prediction using a Cheon-Kim-Kim-Song (CKKS) homomorphic encryption scheme. The proposed hierarchical system model consists of vehicles, roadside units (RSUs), and a central server, enabling collaborative traffic data analysis while maintaining data privacy. The performance of the federated learning system was evaluated with and without the CKKS encryption. In the CKKS-enabled configuration, local models initially exhibit higher errors, with mean squared errors (MSE) ranging from 0.1375 to 0.1621 and negative R² values. However, after ten training rounds, the models show significant improvements, with MSE dropping to 0.0007–0.0008 and R² values exceeding 0.9820, indicating effective learning of traffic patterns. In contrast, the non-encrypted federated learning system demonstrates strong performance from the outset, with MSE values around 0.0007–0.0008 and R² values of 0.98, further improving to an MSE of 0.0003 and R² approaching 0.993 by the end of training. While the non-encrypted system achieves high accuracy, it compromises data privacy. This study highlights the importance of balancing the predictive performance and data protection in privacy-sensitive applications. The proposed framework has practical implications for urban traffic management, enabling real-time data analysis without exposing personal identifiable information. Future research directions include exploring alternative encryption methods, optimizing architectures for faster convergence, and applying the framework to other domains such as smart cities and autonomous vehicles.