Table 3 Effect of number of rounds on accuracy and averaged performance metrics for FL algorithms in highly heterogeneous setting.
From: A comparative study of federated learning methods for COVID-19 detection
Method | Number of rounds and accuracy | Performance metrics (avg. every round) | ||||||
|---|---|---|---|---|---|---|---|---|
3 rounds | 5 rounds | 10 rounds | 15 rounds | Accuracy (%) | Recall (%) | Precision (%) | F1 score (%) | |
FedAVG | 50.17% \(\downarrow\)(− 5.88%) | 54.24% \(\downarrow\)(− 9.54%) | 59.19 \(\downarrow\)(− 10.45%) | 67.04% \(\downarrow\)(− 3.69%) | 58.12 | 60.71 | 42.04 | 45.87 |
FedSGD | 51.30% \(\uparrow\)(+ 0.42%) | 52.29% \(\downarrow\)(− 3.61%) | 60.59% \(\downarrow\)(− 14.99%) | 64.25% \(\downarrow\)(− 12.69%) | 60.09 | 56.97 | 41.44 | 32.18 |
CWT | 49.95% \(\downarrow\)(− 30.82%) | 51.30% \(\downarrow\)(− 38.48%) | 50.84% \(\downarrow\)(− 40.43%) | 54.39% \(\downarrow\)(− 39.17%) | 51.75 | 55.06 | 41.10 | 47.62 |
SWT | – | – | – | – | 48.60 | 47.42 | 34.78 | 49.30 |
STWT | 48.38% \(\downarrow\)(− 42.35%) | 51.65% \(\downarrow\)(− 32.32%) | 50.64% \(\downarrow\)(− 38.80%) | 49.01% \(\downarrow\)(− 43.99%) | 50.91 | 54.02 | 38.64 | 44.71 |
CDS | 85.06% (± 0.00%) | 81.56% (± 0.00%) | 91.06% (± 0.00%) | 91.04% (± 0.00%) | 87.75 | 89.57 | 87.93 | 87.19 |
Local | 43.89% \(\downarrow\)(− 3.26%) | 40.75% \(\downarrow\)(− 11.43%) | 48.98% \(\downarrow\)(− 11.33%) | 47.12% \(\downarrow\)(− 15.68%) | 46.62 | 50.01 | 29.64 | 39.43 |
