Scholarlink

Breast cancer diagnosis represents a pivotal use case for artificial intelligence (AI), necessitating approaches that safeguard data privacy while offering interpretable predictions. This investigation proposes a federated learning (FL) framework enhanced with differential privacy (DP) and explainable AI (XAI) techniques to address these dual challenges. FL enables decentralized model training across multiple clients, preserving the privacy of sensitive healthcare data. DP mechanisms enhance data security, while XAI tools such as Local Interpretable Model-Agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP) offer clarity on model predictions, promoting transparency in clinical decision-making. The framework was evaluated using the Wisconsin Breast Cancer Dataset, employing four classifiers: Stochastic Gradient Descent Classifier(SGDC), Logistic Regression Classifier (LRC), Passive Aggressive Classifier (PAC), and Ridge Classifier (RC). SGDC achieved the highest accuracy ( \(98\%\) ) and the lowest global loss (0.0685), while PAC demonstrated the lowest communication time (0.0057 seconds per CR), making it highly efficient for resource-constrained environments. Feature importance analysis identified radius_mean, area_mean, and perimeter_mean as key contributors across all classifiers. These findings highlight the adaptability of the proposed framework, offering a robust, privacy-preserving, and interpretable solution for breast cancer diagnosis.

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

Enhancing Breast Cancer Detection Through Explainable Federated Learning Models

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Akram Pasha,
  • Syed Ziaur Rahman,
  • Shaik Sayeed Ahamed,
  • D. N. Punith Kumar,
  • R. Manjunatha

摘要

Breast cancer diagnosis represents a pivotal use case for artificial intelligence (AI), necessitating approaches that safeguard data privacy while offering interpretable predictions. This investigation proposes a federated learning (FL) framework enhanced with differential privacy (DP) and explainable AI (XAI) techniques to address these dual challenges. FL enables decentralized model training across multiple clients, preserving the privacy of sensitive healthcare data. DP mechanisms enhance data security, while XAI tools such as Local Interpretable Model-Agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP) offer clarity on model predictions, promoting transparency in clinical decision-making. The framework was evaluated using the Wisconsin Breast Cancer Dataset, employing four classifiers: Stochastic Gradient Descent Classifier(SGDC), Logistic Regression Classifier (LRC), Passive Aggressive Classifier (PAC), and Ridge Classifier (RC). SGDC achieved the highest accuracy ( \(98\%\) ) and the lowest global loss (0.0685), while PAC demonstrated the lowest communication time (0.0057 seconds per CR), making it highly efficient for resource-constrained environments. Feature importance analysis identified radius_mean, area_mean, and perimeter_mean as key contributors across all classifiers. These findings highlight the adaptability of the proposed framework, offering a robust, privacy-preserving, and interpretable solution for breast cancer diagnosis.