Scholarlink

<p>Federated learning is a decentralized approach that enables machine learning models to be trained on local data across distributed clients, offering an alternative to centralized learning by protecting data privacy. Tree-based ensemble methods, such as Random Forest and XGBoost, have demonstrated high performance in various tasks, but their application in federated learning environments especially with Non-IID (Non-Independent and Identically Distributed) data presents challenges in model accuracy and complexity. This paper proposes a federated learning method based on vector-valued XGBoost, specifically designed to address multi-class classification problems in Non-IID data environments. Our method divides the tree-building process into two phases: tree growth and global weight adjustment. These phases are combined to form four types of federated XGBoost models, allowing flexible adjustments in performance and model complexity. Through experiments conducted in federated learning environments, we demonstrate that our proposed method achieves high accuracy while reducing model complexity in Non-IID data settings.</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

Effective federated XGBoost learning for multi-class classification in Non-IID environments

  • 期刊论文

  • 发表日期：

    2025年4月24日

查看全文
  • Sung Won Kang,
  • Cheong Hee Park

摘要

Federated learning is a decentralized approach that enables machine learning models to be trained on local data across distributed clients, offering an alternative to centralized learning by protecting data privacy. Tree-based ensemble methods, such as Random Forest and XGBoost, have demonstrated high performance in various tasks, but their application in federated learning environments especially with Non-IID (Non-Independent and Identically Distributed) data presents challenges in model accuracy and complexity. This paper proposes a federated learning method based on vector-valued XGBoost, specifically designed to address multi-class classification problems in Non-IID data environments. Our method divides the tree-building process into two phases: tree growth and global weight adjustment. These phases are combined to form four types of federated XGBoost models, allowing flexible adjustments in performance and model complexity. Through experiments conducted in federated learning environments, we demonstrate that our proposed method achieves high accuracy while reducing model complexity in Non-IID data settings.