Enhancing Hand Vein Authentication: Gray Wolf Optimization Mitigates Data Poisoning Attacks

Abstract

Nowadays most vital organizations depend on biometric systems to protect their resources and authenticate their customers. Hand vein is one of the biometrics which used for more critical infrastructures. In this paper a hand vein pattern-based biometric authentication system is proposed. The suggested system is designed with attack phase and defense phase. In attack phase, hand veins are preprocessed, and features extracted using PCA-net. These features are clustered using Gray Wolf Optimization (GWO). Lastly, hand veins are poisoned using label flipping mechanism. In defense phase, poisoned hand vein samples are discriminated using Ada-Boost algorithm. The proposed system is highly secure and minimize security loopholes of existing authentication system, through best results: (717) features by PCA-net; best fitness and Silhouette score from GWO (0.75997); ADA-Boost accuracy (baseline step) (0.664857), ADA-Boost accuracy (poisoning step) (0.820512), ADA-Boost accuracy (correction step) (0.827298). Experimental results confirm the effectiveness of the introduced method, where cluster performance improvement is indicated in terms of fitness convergence and silhouette scores. Results show the effectiveness of biometric-based authentication to prevent cyber attacks with privacy and security.