Blind Federated Learning without initial model

Journal of Big Data

Table 10 Blended Blind FL results (accuracy-based weights, slope 5, sigmoid)—Breast Cancer dataset

Agent	Size	% 1s	Accuracy	Accuracy	Precision	Precision
Agent	Size	% 1s	Pre-FL	Post-FL	Pre-FL	Post-FL
1	20%	14%	0.7727	0.9091	0.3750	0.6000
2	20%	26%	0.9130	0.9130	0.8333	0.8333
3	20%	35%	0.7391	0.9130	0.5714	0.7778
4	20%	13%	0.9565	0.9565	0.7500	0.5000
5	20%	26%	0.9565	1.0000	1.0000	1.0000
Avg.	–	–	0.8676	0.9383	0.7060	0.7422
1	5%	27%	0.9063	0.9063	0.6667	0.6667
2	4%	43%	0.9524	0.9524	0.9583	0.8846
3	42%	35%	0.7317	0.7317	1.0000	1.0000
4	20%	23%	0.9231	0.9231	0.7500	0.7500
5	29%	49%	0.8519	0.8704	0.8636	0.9048
Avg.	–	–	0.8731	0.8768	0.8477	0.8412
1	19%	24%	0.9167	0.9167	1.0000	0.8000
2	26%	35%	0.9286	0.9286	1.0000	1.0000
3	16%	42%	0.8990	0.9091	0.9643	0.8750
4	15%	20%	0.9268	0.9268	0.8750	0.8750
5	24%	47%	0.8730	0.8889	0.9565	1.0000
Avg.	–	–	0.9088	0.9140	0.9592	0.9100
1	3%	20%	1.0000	1.0000	1.0000	1.0000
2	4%	36%	0.9231	0.9231	1.0000	1.0000
3	45%	40%	0.8835	0.8932	0.9375	0.9394
4	21%	17%	0.9767	0.9767	0.9167	0.8462
5	26%	49%	0.9167	0.9333	0.9600	1.0000
Avg.	–	–	0.9400	0.9453	0.9628	0.9571