Scholarlink

In the constantly evolving field of cybersecurity, safeguarding sensitive data from hazardous incidents is critical. Traditional intrusion detection systems (IDS) frequently rely on centralized data gathering and processing, which raises privacy and scalability concerns. In this research, we present a privacy-preserving, explainable IDS using Federated Learning (FL) and deep learning techniques. Specifically, we integrate Artificial Neural Networks (ANN) and SHapley Additive exPlanations (SHAP) at the local model level to enhance interpretability. We have also experimented with few ML architecture and a DL architecture to observe how well-structured and decent the dataset is. The technique attempts to improve detection capabilities while protecting data privacy and providing fair decision-making procedures. By utilizing Transformers, known for their superior performance in sequence modeling and anomaly detection, the system can effectively analyze complex network traffic patterns. Our FL-based approach preserves data privacy by training models locally, while collaborative learning improves model robustness. Through experimental analysis, we highlight that while centralized models outperform federated ones in accuracy, the FL model provides significant benefits in terms of privacy and scalability. Our approach emphasizes fair decision-making and interpretability, making it suitable for real-world applications where data confidentiality and transparency are crucial.

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

Securing Networks: A Deep Learning Approach with Explainable AI (XAI) and Federated Learning for Intrusion Detection

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Kazi Fatema,
  • Mehrin Anannya,
  • Samrat Kumar Dey,
  • Chunhua Su,
  • Rashed Mazumder

摘要

In the constantly evolving field of cybersecurity, safeguarding sensitive data from hazardous incidents is critical. Traditional intrusion detection systems (IDS) frequently rely on centralized data gathering and processing, which raises privacy and scalability concerns. In this research, we present a privacy-preserving, explainable IDS using Federated Learning (FL) and deep learning techniques. Specifically, we integrate Artificial Neural Networks (ANN) and SHapley Additive exPlanations (SHAP) at the local model level to enhance interpretability. We have also experimented with few ML architecture and a DL architecture to observe how well-structured and decent the dataset is. The technique attempts to improve detection capabilities while protecting data privacy and providing fair decision-making procedures. By utilizing Transformers, known for their superior performance in sequence modeling and anomaly detection, the system can effectively analyze complex network traffic patterns. Our FL-based approach preserves data privacy by training models locally, while collaborative learning improves model robustness. Through experimental analysis, we highlight that while centralized models outperform federated ones in accuracy, the FL model provides significant benefits in terms of privacy and scalability. Our approach emphasizes fair decision-making and interpretability, making it suitable for real-world applications where data confidentiality and transparency are crucial.