Scholarlink

This study offers a methodology for assessing intrusion detection systems’ (IDS) use of black-box explainable AI (XAI) techniques. Global and local explanation techniques, including LIME and SHAP, are applied to multiple IDS models trained on three datasets: SIMARGL 2021, NSLKDD, and CICIDS 2017. The integrated XAI methods enhance transparency and trust in model decisions. Experimental results show the Voting Classifier (boosted decision trees + bagging random forests) achieves top accuracy: 97.9% on CICIDS 2017, 99.4 on SIMARGL 2021. The results confirm the framework’s effectiveness in combining high detection performance with interpretability.

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

E-XAI: Evaluating Black-Box Explainable AI Frameworks for Network Intrusion Detection

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ankit Kumar,
  • Rishav Sikdar,
  • U. Trisha,
  • Raghavendra Singh Negi,
  • K. M. Yogesh,
  • Vedavati Bhandari

摘要

This study offers a methodology for assessing intrusion detection systems’ (IDS) use of black-box explainable AI (XAI) techniques. Global and local explanation techniques, including LIME and SHAP, are applied to multiple IDS models trained on three datasets: SIMARGL 2021, NSLKDD, and CICIDS 2017. The integrated XAI methods enhance transparency and trust in model decisions. Experimental results show the Voting Classifier (boosted decision trees + bagging random forests) achieves top accuracy: 97.9% on CICIDS 2017, 99.4 on SIMARGL 2021. The results confirm the framework’s effectiveness in combining high detection performance with interpretability.