Fed-Fi: Federated Learning Malicious Model Detection Method Based on Feature Importance

Table 1

Performance index of malicious detection algorithm under sybil attacks.

Malicious detection algorithm		Source: 2, target: 6 (%)	Source: 4, target: 8 (%)
Median	Accuracy	0	0
	False alarm	98.87	99.79
Multi-Krum	Accuracy	0	0
	False alarm	94.57	98.47
Fool Golds	Accuracy	88.27	91.95
	False alarm	0.48	0.92
HMFL	Accuracy	91.57	92.32
	False alarm	0.2	0.31