Skip to content
InfoResearchPeer-reviewed

Adversarial Malware Detection Using Feature Contribution Difference Analysis with Kernel SHAP

Published
Record updated
View JSON

Summary

Researchers propose a two-stage adversarial malware detection method that uses Kernel SHAP to measure each feature's contribution to a neural network classifier's predictions. The method weights features with small benign-versus-malware contribution differences and computes a maliciousness score to redetect adversarial samples misclassified as benign. Experiments on Android application datasets show it significantly outperforms existing approaches against four adversarial attacks (fast gradient sign method, JSMA, Mimicry, and Pointwise) without learning attack-specific characteristics.