Scholarlink

<p>With the rapid growth of computing technologies, edge infrastructure is increasingly facing limitations in computation, bandwidth, and storage. A major challenge in edge computing is the ability to classify data accurately and rapidly without relying on centralized servers. This paper introduces Fed-RFOF, a federated model that combines Random Forest, Ontology, and Fuzzy Logic to enhance prediction accuracy in serverless edge environments. The model comprises four phases: Preprocessing, for initial filtering and outlier removal; Feature Selection, using ontology and fuzzy logic to extract rules and key features; Execution, for local training of decision trees and ensemble creation; and Detection, for local data prediction. The approach is evaluated using DDoS, Botnet, NSL-KDD, and Cyberattack datasets. Simulation results demonstrate that Fed-RFOF achieves an average prediction accuracy of 99.38%, outperforming existing methods by 4.93%. The improvement is statistically significant, with a 95% confidence interval (98.52–100%) and a standard deviation of 0.42%, confirming its effectiveness in decentralized edge computing environments.</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

Improving prediction accuracy in serverless edge computing using a federated learning

  • 期刊论文

  • 发表日期：

    2025年12月29日

查看全文
  • Nasrin Kasiri,
  • Mohammadreza Mollahoseini-Ardakani,
  • Mostafa Ghobaei-Arani

摘要

With the rapid growth of computing technologies, edge infrastructure is increasingly facing limitations in computation, bandwidth, and storage. A major challenge in edge computing is the ability to classify data accurately and rapidly without relying on centralized servers. This paper introduces Fed-RFOF, a federated model that combines Random Forest, Ontology, and Fuzzy Logic to enhance prediction accuracy in serverless edge environments. The model comprises four phases: Preprocessing, for initial filtering and outlier removal; Feature Selection, using ontology and fuzzy logic to extract rules and key features; Execution, for local training of decision trees and ensemble creation; and Detection, for local data prediction. The approach is evaluated using DDoS, Botnet, NSL-KDD, and Cyberattack datasets. Simulation results demonstrate that Fed-RFOF achieves an average prediction accuracy of 99.38%, outperforming existing methods by 4.93%. The improvement is statistically significant, with a 95% confidence interval (98.52–100%) and a standard deviation of 0.42%, confirming its effectiveness in decentralized edge computing environments.