Robustness analysis of post-hoc XAI explanations under adversarial perturbations — covering SHAP, LIME, and Integrated Gradients across phishing, intrusion detection, and fraud datasets, with four attack strategies and hybrid defense achieving 91.1% explanation drift reduction.
mithinsagar/xai-attack-defense-framework: Robustness analysis of post-hoc XAI explanations under adversarial perturbations — covering SHAP, LIME, and Integrated Gradients across phishing, intrusion detection, and fraud datasets, with four attack strategies and hybrid defense achieving 91.1% explanation drift reduction.
