Scholarlink

<p>Federated Learning (FL) is a decentralized machine learning paradigm that enables multiple clients—such as mobile devices, IoT systems and organizational servers—to collaboratively train a global model without exposing their raw local data. This approach ensures data privacy by keeping training data on the client side. Traditional machine learning methods rely on fully labelled datasets, which are often impractical to obtain in real-world scenarios and may lead to privacy concerns. To address these limitations, we propose FedSemdroid, a Federated semi-supervised learning model for Android malware detection. FedSemdroid leverages high-confidence consensus among multiple client models to generate labels, thereby minimizing the need for fully labelled data while preserving data privacy. Experimental results demonstrate that the proposed framework outperforms existing approaches, achieving an accuracy of 98.7%, an F-measure of 0.95, global precision and global recall of 0.97 and 0.96 respectively highlighting its effectiveness and robustness in secure malware detection.</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

FedSemdroid- A Semi-Supervised Algorithm and Framework Based on Federated Learning for Android Malware Detection

  • 期刊论文

  • 发表日期：

    2026年4月9日

查看全文
  • Pooja .,
  • Arvind Mahindru,
  • Pardeep Kumar Arora

摘要

Federated Learning (FL) is a decentralized machine learning paradigm that enables multiple clients—such as mobile devices, IoT systems and organizational servers—to collaboratively train a global model without exposing their raw local data. This approach ensures data privacy by keeping training data on the client side. Traditional machine learning methods rely on fully labelled datasets, which are often impractical to obtain in real-world scenarios and may lead to privacy concerns. To address these limitations, we propose FedSemdroid, a Federated semi-supervised learning model for Android malware detection. FedSemdroid leverages high-confidence consensus among multiple client models to generate labels, thereby minimizing the need for fully labelled data while preserving data privacy. Experimental results demonstrate that the proposed framework outperforms existing approaches, achieving an accuracy of 98.7%, an F-measure of 0.95, global precision and global recall of 0.97 and 0.96 respectively highlighting its effectiveness and robustness in secure malware detection.