Table 24 Robustness analysis on federated learning vs. Centralized.

From: A federated learning-based privacy-preserving image processing framework for brain tumor detection from CT scans

Scenario

Accuracy

F1-score

Dropout recovery (%)

Resilience score*

No failures (ideal FL)

98.32%

98.22%

100%

0.98

10% client dropout

97.40%

97.01%

96%

0.93

10% adversarial noise (model poisoning)

96.12%

95.88%

92%

0.91

Centralized (reference)

99.07%

99.55%

N/A

1

Back to article page